fiftyone.core.collections

Interface for sample collections.

Copyright 2017-2025, Voxel51, Inc.

voxel51.com

Functions:

view_stage(func)
aggregation(func)

Classes:

SaveContext(sample_collection[, batch_size, ...])	Context that saves samples from a collection according to a configurable batching strategy.
SampleCollection()	Abstract class representing an ordered collection of fiftyone.core.sample.Sample instances in a fiftyone.core.dataset.Dataset.

fiftyone.core.collections.view_stage(func)

fiftyone.core.collections.aggregation(func)

class fiftyone.core.collections.SaveContext(sample_collection, batch_size=None, batching_strategy=None)

Bases: object

Context that saves samples from a collection according to a configurable batching strategy.

Parameters:

• sample_collection – a fiftyone.core.collections.SampleCollection

• batch_size (None) – the batch size to use. If a batching_strategy is provided, this parameter configures the strategy as described below. If no batching_strategy is provided, this can either be an integer specifying the number of samples to save in a batch (in which case batching_strategy is implicitly set to "static") or a float number of seconds between batched saves (in which case batching_strategy is implicitly set to "latency")

• batching_strategy (None) –

  the batching strategy to use for each save operation. Supported values are:

  • "static": a fixed sample batch size for each save

  • "size": a target batch size, in bytes, for each save

  • "latency": a target latency, in seconds, between saves

  By default, fo.config.default_batcher is used

Methods:

save(sample)

Registers the sample for saving in the next batch.

save(sample)

Registers the sample for saving in the next batch.

Parameters:

sample – a fiftyone.core.sample.Sample or fiftyone.core.sample.SampleView

class fiftyone.core.collections.SampleCollection

Bases: object

Abstract class representing an ordered collection of fiftyone.core.sample.Sample instances in a fiftyone.core.dataset.Dataset.

Attributes:

name	The name of the collection.
media_type	The media type of the collection.
group_field	The group field of the collection, or None if the collection is not grouped.
group_slice	The current group slice of the collection, or None if the collection is not grouped.
group_slices	The list of group slices of the collection, or None if the collection is not grouped.
group_media_types	A dict mapping group slices to media types, or None if the collection is not grouped.
default_group_slice	The default group slice of the collection, or None if the collection is not grouped.
tags	The list of tags of the underlying dataset.
description	A description of the underlying dataset.
info	The info dict of the underlying dataset.
app_config	Dataset-specific settings that customize how this collection is visualized in the FiftyOne App.
classes	The classes of the underlying dataset.
default_classes	The default classes of the underlying dataset.
mask_targets	The mask targets of the underlying dataset.
default_mask_targets	The default mask targets of the underlying dataset.
skeletons	The keypoint skeletons of the underlying dataset.
default_skeleton	The default keypoint skeleton of the underlying dataset.
has_evaluations	Whether this collection has any evaluation results.
has_brain_runs	Whether this collection has any brain runs.
has_runs	Whether this collection has any runs.
has_annotation_runs	Whether this collection has any annotation runs.

Methods:

has_classes(field)	Determines whether this collection has a classes list for the given field.
get_classes(field)	Gets the classes list for the given field, or None if no classes are available.
has_mask_targets(field)	Determines whether this collection has mask targets for the given field.
get_mask_targets(field)	Gets the mask targets for the given field, or None if no mask targets are available.
has_skeleton(field)	Determines whether this collection has a keypoint skeleton for the given field.
get_skeleton(field)	Gets the keypoint skeleton for the given field, or None if no skeleton is available.
summary()	Returns a string summary of the collection.
sync_last_modified_at([include_frames])	Syncs the last_modified_at property(s) of the dataset.
stats([include_media, include_indexes, ...])	Returns stats about the collection on disk.
first()	Returns the first sample in the collection.
last()	Returns the last sample in the collection.
head([num_samples])	Returns a list of the first few samples in the collection.
tail([num_samples])	Returns a list of the last few samples in the collection.
one(expr[, exact])	Returns a single sample in this collection matching the expression.
view()	Returns a fiftyone.core.view.DatasetView containing the collection.
iter_samples([progress, autosave, ...])	Returns an iterator over the samples in the collection.
iter_groups([group_slices, progress, ...])	Returns an iterator over the groups in the collection.
get_group(group_id[, group_slices])	Returns a dict containing the samples for the given group ID.
save_context([batch_size, batching_strategy])	Returns a context that can be used to save samples from this collection according to a configurable batching strategy.
get_field(path[, ftype, embedded_doc_type, ...])	Returns the field instance of the provided path, or None if one does not exist.
get_field_schema([ftype, embedded_doc_type, ...])	Returns a schema dictionary describing the fields of the samples in the collection.
get_frame_field_schema([ftype, ...])	Returns a schema dictionary describing the fields of the frames in the collection.
get_dynamic_field_schema([fields, recursive])	Returns a schema dictionary describing the dynamic fields of the samples in the collection.
get_dynamic_frame_field_schema([fields, ...])	Returns a schema dictionary describing the dynamic fields of the frames in the collection.
make_unique_field_name([root])	Makes a unique field name with the given root name for the collection.
has_field(path)	Determines whether the collection has a field with the given name.
has_sample_field(path)	Determines whether the collection has a sample field with the given name.
has_frame_field(path)	Determines whether the collection has a frame-level field with the given name.
validate_fields_exist(fields[, include_private])	Validates that the collection has field(s) with the given name(s).
validate_field_type(path[, ftype, ...])	Validates that the collection has a field of the given type.
tag_samples(tags)	Adds the tag(s) to all samples in this collection, if necessary.
untag_samples(tags)	Removes the tag(s) from all samples in this collection, if necessary.
count_sample_tags()	Counts the occurrences of sample tags in this collection.
tag_labels(tags[, label_fields])	Adds the tag(s) to all labels in the specified label field(s) of this collection, if necessary.
untag_labels(tags[, label_fields])	Removes the tag from all labels in the specified label field(s) of this collection, if necessary.
count_label_tags([label_fields])	Counts the occurrences of all label tags in the specified label field(s) of this collection.
split_labels(in_field, out_field[, filter])	Splits the labels from the given input field into the given output field of the collection.
merge_labels(in_field, out_field)	Merges the labels from the given input field into the given output field of the collection.
set_values(field_name, values[, key_field, ...])	Sets the field or embedded field on each sample or frame in the collection to the given values.
set_label_values(field_name, values[, ...])	Sets the fields of the specified labels in the collection to the given values.
compute_metadata([overwrite, num_workers, ...])	Populates the metadata field of all samples in the collection.
apply_model(model[, label_field, ...])	Applies the model to the samples in the collection.
compute_embeddings(model[, ...])	Computes embeddings for the samples in the collection using the given model.
compute_patch_embeddings(model, patches_field)	Computes embeddings for the image patches defined by patches_field of the samples in the collection using the given model.
evaluate_regressions(pred_field[, gt_field, ...])	Evaluates the regression predictions in this collection with respect to the specified ground truth values.
evaluate_classifications(pred_field[, ...])	Evaluates the classification predictions in this collection with respect to the specified ground truth labels.
evaluate_detections(pred_field[, gt_field, ...])	Evaluates the specified predicted detections in this collection with respect to the specified ground truth detections.
evaluate_segmentations(pred_field[, ...])	Evaluates the specified semantic segmentation masks in this collection with respect to the specified ground truth masks.
has_evaluation(eval_key)	Whether this collection has an evaluation with the given key.
list_evaluations([type, method])	Returns a list of evaluation keys on this collection.
map_samples(map_fcn[, save, skip_failures, ...])	Applies the given function to each sample in the collection and returns the results as a generator.
update_samples(update_fcn[, skip_failures, ...])	Applies the given function to each sample in the collection and saves the resulting sample edits.
rename_evaluation(eval_key, new_eval_key)	Replaces the key for the given evaluation with a new key.
get_evaluation_info(eval_key)	Returns information about the evaluation with the given key on this collection.
load_evaluation_results(eval_key[, cache])	Loads the results for the evaluation with the given key on this collection.
load_evaluation_view(eval_key[, select_fields])	Loads the fiftyone.core.view.DatasetView on which the specified evaluation was performed on this collection.
delete_evaluation(eval_key)	Deletes the evaluation results associated with the given evaluation key from this collection.
delete_evaluations()	Deletes all evaluation results from this collection.
has_brain_run(brain_key)	Whether this collection has a brain method run with the given key.
list_brain_runs([type, method])	Returns a list of brain keys on this collection.
rename_brain_run(brain_key, new_brain_key)	Replaces the key for the given brain run with a new key.
get_brain_info(brain_key)	Returns information about the brain method run with the given key on this collection.
load_brain_results(brain_key[, cache, load_view])	Loads the results for the brain method run with the given key on this collection.
load_brain_view(brain_key[, select_fields])	Loads the fiftyone.core.view.DatasetView on which the specified brain method run was performed on this collection.
delete_brain_run(brain_key)	Deletes the brain method run with the given key from this collection.
delete_brain_runs()	Deletes all brain method runs from this collection.
has_run(run_key)	Whether this collection has a run with the given key.
list_runs(**kwargs)	Returns a list of run keys on this collection.
init_run(**kwargs)	Initializes a config instance for a new run.
register_run(run_key, config[, results, ...])	Registers a run under the given key on this collection.
rename_run(run_key, new_run_key)	Replaces the key for the given run with a new key.
get_run_info(run_key)	Returns information about the run with the given key on this collection.
update_run_config(run_key, config)	Updates the run config for the run with the given key.
init_run_results(run_key, **kwargs)	Initializes a results instance for the run with the given key.
save_run_results(run_key, results[, ...])	Saves run results for the run with the given key.
load_run_results(run_key[, cache, load_view])	Loads the results for the run with the given key on this collection.
load_run_view(run_key[, select_fields])	Loads the fiftyone.core.view.DatasetView on which the specified run was performed on this collection.
delete_run(run_key)	Deletes the run with the given key from this collection.
delete_runs()	Deletes all runs from this collection.
list_view_stages()	Returns a list of all available methods on this collection that apply fiftyone.core.stages.ViewStage operations to this collection.
add_stage(stage)	Applies the given fiftyone.core.stages.ViewStage to the collection.
concat(samples)	Concatenates the contents of the given SampleCollection to this collection.
exclude(sample_ids)	Excludes the samples with the given IDs from the collection.
exclude_by(field, values)	Excludes the samples with the given field values from the collection.
exclude_fields([field_names, meta_filter, ...])	Excludes the fields with the given names from the samples in the collection.
exclude_frames(frame_ids[, omit_empty])	Excludes the frames with the given IDs from the video collection.
exclude_groups(group_ids)	Excludes the groups with the given IDs from the grouped collection.
exclude_group_slices([slices, media_type])	Excludes the specified group slice(s) from the grouped collection.
exclude_labels([labels, ids, instance_ids, ...])	Excludes the specified labels from the collection.
exists(field[, bool])	Returns a view containing the samples in the collection that have (or do not have) a non-None value for the given field or embedded field.
filter_field(field, filter[, only_matches])	Filters the values of a field or embedded field of each sample in the collection.
filter_labels(field, filter[, only_matches, ...])	Filters the fiftyone.core.labels.Label field of each sample in the collection.
filter_keypoints(field[, filter, labels, ...])	Filters the individual fiftyone.core.labels.Keypoint.points elements in the specified keypoints field of each sample in the collection.
flatten([stages])	Returns a flattened view that contains all samples in the dynamic grouped collection.
geo_near(point[, location_field, ...])	Sorts the samples in the collection by their proximity to a specified geolocation.
geo_within(boundary[, location_field, ...])	Filters the samples in this collection to only include samples whose geolocation is within a specified boundary.
group_by(field_or_expr[, order_by, reverse, ...])	Creates a view that groups the samples in the collection by a specified field or expression.
limit(limit)	Returns a view with at most the given number of samples.
limit_labels(field, limit)	Limits the number of fiftyone.core.labels.Label instances in the specified labels list field of each sample in the collection.
map_labels(field, map)	Maps the label values of a fiftyone.core.labels.Label field to new values for each sample in the collection.
map_values(field, map)	Maps the values in the given field to new values for each sample in the collection.
set_field(field, expr[, _allow_missing])	Sets a field or embedded field on each sample in a collection by evaluating the given expression.
match(filter)	Filters the samples in the collection by the given filter.
match_frames(filter[, omit_empty])	Filters the frames in the video collection by the given filter.
match_labels([labels, ids, instance_ids, ...])	Selects the samples from the collection that contain (or do not contain) at least one label that matches the specified criteria.
match_tags(tags[, bool, all])	Returns a view containing the samples in the collection that have or don't have any/all of the given tag(s).
mongo(pipeline[, _needs_frames, _group_slices])	Adds a view stage defined by a raw MongoDB aggregation pipeline.
select(sample_ids[, ordered])	Selects the samples with the given IDs from the collection.
select_by(field, values[, ordered])	Selects the samples with the given field values from the collection.
select_fields([field_names, meta_filter, ...])	Selects only the fields with the given names from the samples in the collection.
select_frames(frame_ids[, omit_empty])	Selects the frames with the given IDs from the video collection.
select_groups(group_ids[, ordered])	Selects the groups with the given IDs from the grouped collection.
select_group_slices([slices, media_type, ...])	Selects the specified group slice(s) from the grouped collection.
select_labels([labels, ids, instance_ids, ...])	Selects only the specified labels from the collection.
shuffle([seed])	Randomly shuffles the samples in the collection.
skip(skip)	Omits the given number of samples from the head of the collection.
sort_by(field_or_expr[, reverse, create_index])	Sorts the samples in the collection by the given field(s) or expression(s).
sort_by_similarity(query[, k, reverse, ...])	Sorts the collection by similarity to a specified query.
take(size[, seed])	Randomly samples the given number of samples from the collection.
to_patches(field, **kwargs)	Creates a view that contains one sample per object patch in the specified field of the collection.
to_evaluation_patches(eval_key, **kwargs)	Creates a view based on the results of the evaluation with the given key that contains one sample for each true positive, false positive, and false negative example in the collection, respectively.
to_clips(field_or_expr, **kwargs)	Creates a view that contains one sample per clip defined by the given field or expression in the video collection.
to_trajectories(field, **kwargs)	Creates a view that contains one clip for each unique object trajectory defined by their (label, index) in a frame-level field of a video collection.
to_frames(**kwargs)	Creates a view that contains one sample per frame in the video collection.
list_aggregations()	Returns a list of all available methods on this collection that apply fiftyone.core.aggregations.Aggregation operations to this collection.
bounds(field_or_expr[, expr, safe])	Computes the bounds of a numeric field of the collection.
count([field_or_expr, expr, safe])	Counts the number of field values in the collection.
count_values(field_or_expr[, expr, safe])	Counts the occurrences of field values in the collection.
distinct(field_or_expr[, expr, safe])	Computes the distinct values of a field in the collection.
histogram_values(field_or_expr[, expr, ...])	Computes a histogram of the field values in the collection.
min(field_or_expr[, expr, safe])	Computes the minimum of a numeric field of the collection.
max(field_or_expr[, expr, safe])	Computes the maximum of a numeric field of the collection.
mean(field_or_expr[, expr, safe])	Computes the arithmetic mean of the field values of the collection.
quantiles(field_or_expr, quantiles[, expr, safe])	Computes the quantile(s) of the field values of a collection.
schema(field_or_expr[, expr, dynamic_only, ...])	Extracts the names and types of the attributes of a specified embedded document field across all samples in the collection.
list_schema(field_or_expr[, expr])	Extracts the value type(s) in a specified list field across all samples in the collection.
std(field_or_expr[, expr, safe, sample])	Computes the standard deviation of the field values of the collection.
sum(field_or_expr[, expr, safe])	Computes the sum of the field values of the collection.
values(field_or_expr[, expr, missing_value, ...])	Extracts the values of a field from all samples in the collection.
draw_labels(output_dir[, rel_dir, ...])	Renders annotated versions of the media in the collection with the specified label data overlaid to the given directory.
export([export_dir, dataset_type, ...])	Exports the samples in the collection to disk.
to_torch(get_item[, vectorize, ...])	Constructs a torch.utils.data.Dataset that loads data from this collection via the provided fiftyone.utils.torch.GetItem instance.
annotate(anno_key[, label_schema, ...])	Exports the samples and optional label field(s) in this collection to the given annotation backend.
has_annotation_run(anno_key)	Whether this collection has an annotation run with the given key.
list_annotation_runs([type, method])	Returns a list of annotation keys on this collection.
rename_annotation_run(anno_key, new_anno_key)	Replaces the key for the given annotation run with a new key.
get_annotation_info(anno_key)	Returns information about the annotation run with the given key on this collection.
load_annotation_results(anno_key[, cache])	Loads the results for the annotation run with the given key on this collection.
load_annotation_view(anno_key[, select_fields])	Loads the fiftyone.core.view.DatasetView on which the specified annotation run was performed on this collection.
load_annotations(anno_key[, dest_field, ...])	Downloads the labels from the given annotation run from the annotation backend and merges them into this collection.
delete_annotation_run(anno_key)	Deletes the annotation run with the given key from this collection.
delete_annotation_runs()	Deletes all annotation runs from this collection.
list_indexes()	Returns the list of index names on this collection.
get_index_information([include_stats, ...])	Returns a dictionary of information about the indexes on this collection.
create_index(field_or_spec[, unique, wait])	Creates an index on the given field or with the given specification, if necessary.
drop_index(field_or_name)	Drops the index for the given field or name, if necessary.
reload()	Reloads the collection from the database.
to_dict([rel_dir, include_private, ...])	Returns a JSON dictionary representation of the collection.
to_json([rel_dir, include_private, ...])	Returns a JSON string representation of the collection.
write_json(json_path[, rel_dir, ...])	Writes the colllection to disk in JSON format.
aggregate(aggregations[, _mongo])	Aggregates one or more fiftyone.core.aggregations.Aggregation instances.

property name

The name of the collection.

property media_type

The media type of the collection.

property group_field

The group field of the collection, or None if the collection is not grouped.

property group_slice

The current group slice of the collection, or None if the collection is not grouped.

property group_slices

The list of group slices of the collection, or None if the collection is not grouped.

property group_media_types

A dict mapping group slices to media types, or None if the collection is not grouped.

property default_group_slice

The default group slice of the collection, or None if the collection is not grouped.

property tags

The list of tags of the underlying dataset.

See fiftyone.core.dataset.Dataset.tags() for more information.

property description

A description of the underlying dataset.

See fiftyone.core.dataset.Dataset.description() for more information.

property info

The info dict of the underlying dataset.

See fiftyone.core.dataset.Dataset.info() for more information.

property app_config

Dataset-specific settings that customize how this collection is visualized in the FiftyOne App.

property classes

The classes of the underlying dataset.

See fiftyone.core.dataset.Dataset.classes() for more information.

property default_classes

The default classes of the underlying dataset.

See fiftyone.core.dataset.Dataset.default_classes() for more information.

has_classes(field)

Determines whether this collection has a classes list for the given field.

Classes may be defined either in classes() or default_classes().

Parameters:

field – a field name

Returns:

True/False

get_classes(field)

Gets the classes list for the given field, or None if no classes are available.

Classes are first retrieved from classes() if they exist, otherwise from default_classes().

Parameters:

field – a field name

Returns:

a list of classes, or None

property mask_targets

The mask targets of the underlying dataset.

See fiftyone.core.dataset.Dataset.mask_targets() for more information.

property default_mask_targets

The default mask targets of the underlying dataset.

See fiftyone.core.dataset.Dataset.default_mask_targets() for more information.

has_mask_targets(field)

Determines whether this collection has mask targets for the given field.

Mask targets may be defined either in mask_targets() or default_mask_targets().

Parameters:

field – a field name

Returns:

True/False

get_mask_targets(field)

Gets the mask targets for the given field, or None if no mask targets are available.

Mask targets are first retrieved from mask_targets() if they exist, otherwise from default_mask_targets().

Parameters:

field – a field name

Returns:

a list of classes, or None

property skeletons

The keypoint skeletons of the underlying dataset.

See fiftyone.core.dataset.Dataset.skeletons() for more information.

property default_skeleton

The default keypoint skeleton of the underlying dataset.

See fiftyone.core.dataset.Dataset.default_skeleton() for more information.

has_skeleton(field)

Determines whether this collection has a keypoint skeleton for the given field.

Keypoint skeletons may be defined either in skeletons() or default_skeleton().

Parameters:

field – a field name

Returns:

True/False

get_skeleton(field)

Gets the keypoint skeleton for the given field, or None if no skeleton is available.

Skeletons are first retrieved from skeletons() if they exist, otherwise from default_skeleton().

Parameters:

field – a field name

Returns:

a list of classes, or None

summary()

Returns a string summary of the collection.

Returns:

a string summary

sync_last_modified_at(include_frames=True)

Syncs the last_modified_at property(s) of the dataset.

Updates the last_modified_at property of the dataset if necessary to incorporate any modification/deletion timestamps to its samples.

If include_frames==True, the last_modified_at property of each video sample is first updated if necessary to incorporate any modification timestamps to its frames.

Parameters:

include_frames (True) – whether to update the last_modified_at property of video samples. Only applicable to datasets that contain videos

stats(include_media=False, include_indexes=False, compressed=False)

Returns stats about the collection on disk.

The samples keys refer to the sample documents stored in the database.

For video datasets, the frames keys refer to the frame documents stored in the database.

The media keys refer to the raw media associated with each sample on disk.

The index[es] keys refer to the indexes associated with the dataset.

Note that dataset-level metadata such as annotation runs are not included in this computation.

Parameters:

• include_media (False) – whether to include stats about the size of the raw media in the collection

• include_indexes (False) – whether to include stats on the dataset’s indexes

• compressed (False) – whether to return the sizes of collections in their compressed form on disk (True) or the logical uncompressed size of the collections (False). This option is only supported for datasets (not views)

Returns:

a stats dict

first()

Returns the first sample in the collection.

Returns:

a fiftyone.core.sample.Sample or fiftyone.core.sample.SampleView

last()

Returns the last sample in the collection.

Returns:

a fiftyone.core.sample.Sample or fiftyone.core.sample.SampleView

head(num_samples=3)

Returns a list of the first few samples in the collection.

If fewer than num_samples samples are in the collection, only the available samples are returned.

Parameters:

num_samples (3) – the number of samples

Returns:

a list of fiftyone.core.sample.Sample objects

tail(num_samples=3)

Returns a list of the last few samples in the collection.

If fewer than num_samples samples are in the collection, only the available samples are returned.

Parameters:

num_samples (3) – the number of samples

Returns:

a list of fiftyone.core.sample.Sample objects

one(expr, exact=False)

Returns a single sample in this collection matching the expression.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz
from fiftyone import ViewField as F

dataset = foz.load_zoo_dataset("quickstart")

#
# Get a sample by filepath
#

# A random filepath in the dataset
filepath = dataset.take(1).first().filepath

# Get sample by filepath
sample = dataset.one(F("filepath") == filepath)

#
# Dealing with multiple matches
#

# Get a sample whose image is JPEG
sample = dataset.one(F("filepath").ends_with(".jpg"))

# Raises an error since there are multiple JPEGs
dataset.one(F("filepath").ends_with(".jpg"), exact=True)

Parameters:

• expr – a fiftyone.core.expressions.ViewExpression or MongoDB expression that evaluates to True for the sample to match

• exact (False) – whether to raise an error if multiple samples match the expression

Raises:

• ValueError – if no samples match the expression or if exact=True

• and multiple samples match the expression –

Returns:

a fiftyone.core.sample.SampleView

view()

Returns a fiftyone.core.view.DatasetView containing the collection.

Returns:

a fiftyone.core.view.DatasetView

iter_samples(progress=False, autosave=False, batch_size=None, batching_strategy=None)

Returns an iterator over the samples in the collection.

Parameters:

• progress (False) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead

• autosave (False) – whether to automatically save changes to samples emitted by this iterator

• batch_size (None) – the batch size to use when autosaving samples. If a batching_strategy is provided, this parameter configures the strategy as described below. If no batching_strategy is provided, this can either be an integer specifying the number of samples to save in a batch (in which case batching_strategy is implicitly set to "static") or a float number of seconds between batched saves (in which case batching_strategy is implicitly set to "latency")

• batching_strategy (None) –

  the batching strategy to use for each save operation when autosaving samples. Supported values are:

  • "static": a fixed sample batch size for each save

  • "size": a target batch size, in bytes, for each save

  • "latency": a target latency, in seconds, between saves

  By default, fo.config.default_batcher is used

Returns:

an iterator over fiftyone.core.sample.Sample or fiftyone.core.sample.SampleView instances

iter_groups(group_slices=None, progress=False, autosave=False, batch_size=None, batching_strategy=None)

Returns an iterator over the groups in the collection.

Parameters:

• group_slices (None) – an optional subset of group slices to load

• progress (False) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead

• autosave (False) – whether to automatically save changes to samples emitted by this iterator

• batch_size (None) – the batch size to use when autosaving samples. If a batching_strategy is provided, this parameter configures the strategy as described below. If no batching_strategy is provided, this can either be an integer specifying the number of samples to save in a batch (in which case batching_strategy is implicitly set to "static") or a float number of seconds between batched saves (in which case batching_strategy is implicitly set to "latency")

• batching_strategy (None) –

  the batching strategy to use for each save operation when autosaving samples. Supported values are:

  • "static": a fixed sample batch size for each save

  • "size": a target batch size, in bytes, for each save

  • "latency": a target latency, in seconds, between saves

  By default, fo.config.default_batcher is used

Returns:

an iterator that emits dicts mapping group slice names to fiftyone.core.sample.Sample or fiftyone.core.sample.SampleView instances, one per group

get_group(group_id, group_slices=None)

Returns a dict containing the samples for the given group ID.

Parameters:

• group_id – a group ID

• group_slices (None) – an optional subset of group slices to load

Returns:

a dict mapping group names to fiftyone.core.sample.Sample or fiftyone.core.sample.SampleView instances

Raises:

KeyError – if the group ID is not found

save_context(batch_size=None, batching_strategy=None)

Returns a context that can be used to save samples from this collection according to a configurable batching strategy.

Examples:

import random as r
import string as s

import fiftyone as fo
import fiftyone.zoo as foz

dataset = foz.load_zoo_dataset("cifar10", split="test")

def make_label():
    return "".join(r.choice(s.ascii_letters) for i in range(10))

# No save context
for sample in dataset.iter_samples(progress=True):
    sample.ground_truth.label = make_label()
    sample.save()

# Save using default batching strategy
with dataset.save_context() as context:
    for sample in dataset.iter_samples(progress=True):
        sample.ground_truth.label = make_label()
        context.save(sample)

# Save in batches of 10
with dataset.save_context(batch_size=10) as context:
    for sample in dataset.iter_samples(progress=True):
        sample.ground_truth.label = make_label()
        context.save(sample)

# Save every 0.5 seconds
with dataset.save_context(batch_size=0.5) as context:
    for sample in dataset.iter_samples(progress=True):
        sample.ground_truth.label = make_label()
        context.save(sample)

Parameters:

• batch_size (None) – the batch size to use. If a batching_strategy is provided, this parameter configures the strategy as described below. If no batching_strategy is provided, this can either be an integer specifying the number of samples to save in a batch (in which case batching_strategy is implicitly set to "static") or a float number of seconds between batched saves (in which case batching_strategy is implicitly set to "latency")

• batching_strategy (None) –

  the batching strategy to use for each save operation. Supported values are:

  • "static": a fixed sample batch size for each save

  • "size": a target batch size, in bytes, for each save

  • "latency": a target latency, in seconds, between saves

  By default, fo.config.default_batcher is used

Returns:

a SaveContext

get_field(path, ftype=None, embedded_doc_type=None, subfield=None, read_only=None, include_private=False, leaf=False)

Returns the field instance of the provided path, or None if one does not exist.

Parameters:

• path – a field path

• ftype (None) – an optional field type or iterable of types to enforce. Must be subclass(es) of fiftyone.core.fields.Field

• embedded_doc_type (None) – an optional embedded document type or iterable of types to enforce. Must be subclass(es) of fiftyone.core.odm.BaseEmbeddedDocument

• subfield (None) – an optional subfield type or iterable of subfield types to enforce. Must be subclass(es) of fiftyone.core.fields.Field

• read_only (None) – whether to optionally enforce that the field is read-only (True) or not read-only (False)

• include_private (False) – whether to include fields that start with _ in the returned schema

• leaf (False) – whether to return the subfield of list fields

Returns:

a fiftyone.core.fields.Field instance or None

Raises:

ValueError – if the field does not match provided constraints

get_field_schema(ftype=None, embedded_doc_type=None, subfield=None, read_only=None, info_keys=None, created_after=None, include_private=False, flat=False, unwind=True, mode=None)

Returns a schema dictionary describing the fields of the samples in the collection.

Parameters:

• ftype (None) – an optional field type or iterable of types to which to restrict the returned schema. Must be subclass(es) of fiftyone.core.fields.Field

• embedded_doc_type (None) – an optional embedded document type or iterable of types to which to restrict the returned schema. Must be subclass(es) of fiftyone.core.odm.BaseEmbeddedDocument

• subfield (None) – an optional subfield type or iterable of subfield types to which to restrict the returned schema. Must be subclass(es) of fiftyone.core.fields.Field

• read_only (None) – whether to restrict to (True) or exclude (False) read-only fields. By default, all fields are included

• info_keys (None) – an optional key or list of keys that must be in the field’s info dict

• created_after (None) – an optional datetime specifying a minimum creation date

• include_private (False) – whether to include fields that start with _ in the returned schema

• flat (False) – whether to return a flattened schema where all embedded document fields are included as top-level keys

• unwind (True) – whether to traverse into list fields. Only applicable when flat=True

• mode (None) – whether to apply the above constraints before and/or after flattening the schema. Only applicable when flat=True. Supported values are ("before", "after", "both"). The default is "after"

Returns:

a dict mapping field names to fiftyone.core.fields.Field instances

get_frame_field_schema(ftype=None, embedded_doc_type=None, subfield=None, read_only=None, info_keys=None, created_after=None, include_private=False, flat=False, unwind=True, mode=None)

Returns a schema dictionary describing the fields of the frames in the collection.

Only applicable for collections that contain videos.

Parameters:

• ftype (None) – an optional field type to which to restrict the returned schema. Must be a subclass of fiftyone.core.fields.Field

• embedded_doc_type (None) – an optional embedded document type to which to restrict the returned schema. Must be a subclass of fiftyone.core.odm.BaseEmbeddedDocument

• subfield (None) – an optional subfield type or iterable of subfield types to which to restrict the returned schema. Must be subclass(es) of fiftyone.core.fields.Field

• read_only (None) – whether to restrict to (True) or exclude (False) read-only fields. By default, all fields are included

• info_keys (None) – an optional key or list of keys that must be in the field’s info dict

• created_after (None) – an optional datetime specifying a minimum creation date

• include_private (False) – whether to include fields that start with _ in the returned schema

• flat (False) – whether to return a flattened schema where all embedded document fields are included as top-level keys

• unwind (True) – whether to traverse into list fields. Only applicable when flat=True

• mode (None) – whether to apply the above constraints before and/or after flattening the schema. Only applicable when flat=True. Supported values are ("before", "after", "both"). The default is "after"

Returns:

a dict mapping field names to fiftyone.core.fields.Field instances, or None if the collection does not contain videos

get_dynamic_field_schema(fields=None, recursive=True)

Returns a schema dictionary describing the dynamic fields of the samples in the collection.

Dynamic fields are embedded document fields with at least one non-None value that have not been declared on the dataset’s schema.

Parameters:

• fields (None) – an optional field or iterable of fields for which to return dynamic fields. By default, all fields are considered

• recursive (True) – whether to recursively inspect nested lists and embedded documents

Returns:

a dict mapping field paths to fiftyone.core.fields.Field instances or lists of them

get_dynamic_frame_field_schema(fields=None, recursive=True)

Returns a schema dictionary describing the dynamic fields of the frames in the collection.

Dynamic fields are embedded document fields with at least one non-None value that have not been declared on the dataset’s schema.

Parameters:

• fields (None) – an optional field or iterable of fields for which to return dynamic fields. By default, all fields are considered

• recursive (True) – whether to recursively inspect nested lists and embedded documents

Returns:

a dict mapping field paths to fiftyone.core.fields.Field instances or lists of them, or None if the collection does not contain videos

make_unique_field_name(root='')

Makes a unique field name with the given root name for the collection.

Parameters:

root – an optional root for the output field name

Returns:

the field name

has_field(path)

Determines whether the collection has a field with the given name.

Parameters:

path – the field name or embedded.field.name

Returns:

True/False

has_sample_field(path)

Determines whether the collection has a sample field with the given name.

Parameters:

path – the field name or embedded.field.name

Returns:

True/False

has_frame_field(path)

Determines whether the collection has a frame-level field with the given name.

Parameters:

path – the field name or embedded.field.name

Returns:

True/False

validate_fields_exist(fields, include_private=False)

Validates that the collection has field(s) with the given name(s).

If embedded field names are provided, only the root field is checked.

Parameters:

• fields – a field name or iterable of field names

• include_private (False) – whether to include private fields when checking for existence

Raises:

ValueError – if one or more of the fields do not exist

validate_field_type(path, ftype=None, embedded_doc_type=None, subfield=None)

Validates that the collection has a field of the given type.

Parameters:

• path – a field name or embedded.field.name

• ftype (None) – an optional field type or iterable of types to enforce. Must be subclass(es) of fiftyone.core.fields.Field

• embedded_doc_type (None) – an optional embedded document type or iterable of types to enforce. Must be subclass(es) of fiftyone.core.odm.BaseEmbeddedDocument

• subfield (None) – an optional subfield type or iterable of subfield types to enforce. Must be subclass(es) of fiftyone.core.fields.Field

Raises:

ValueError – if the field does not exist or does not have the expected type

tag_samples(tags)

Adds the tag(s) to all samples in this collection, if necessary.

Parameters:

tags – a tag or iterable of tags

untag_samples(tags)

Removes the tag(s) from all samples in this collection, if necessary.

Parameters:

tags – a tag or iterable of tags

count_sample_tags()

Counts the occurrences of sample tags in this collection.

Returns:

a dict mapping tags to counts

tag_labels(tags, label_fields=None)

Adds the tag(s) to all labels in the specified label field(s) of this collection, if necessary.

Parameters:

• tags – a tag or iterable of tags

• label_fields (None) – an optional name or iterable of names of fiftyone.core.labels.Label fields. By default, all label fields are used

untag_labels(tags, label_fields=None)

Removes the tag from all labels in the specified label field(s) of this collection, if necessary.

Parameters:

• tags – a tag or iterable of tags

• label_fields (None) – an optional name or iterable of names of fiftyone.core.labels.Label fields. By default, all label fields are used

count_label_tags(label_fields=None)

Counts the occurrences of all label tags in the specified label field(s) of this collection.

Parameters:

label_fields (None) – an optional name or iterable of names of fiftyone.core.labels.Label fields. By default, all label fields are used

Returns:

a dict mapping tags to counts

split_labels(in_field, out_field, filter=None)

Splits the labels from the given input field into the given output field of the collection.

This method is typically invoked on a view that has filtered the contents of the specified input field, so that the labels in the view are moved to the output field and the remaining labels are left in-place.

Alternatively, you can provide a filter expression that selects the labels of interest to move in this collection.

Parameters:

• in_field – the name of the input label field

• out_field – the name of the output label field, which will be created if necessary

• filter (None) – a boolean fiftyone.core.expressions.ViewExpression to apply to each label in the input field to determine whether to move it (True) or leave it (False)

merge_labels(in_field, out_field)

Merges the labels from the given input field into the given output field of the collection.

If this collection is a dataset, the input field is deleted after the merge.

If this collection is a view, the input field will still exist on the underlying dataset but will only contain the labels not present in this view.

Parameters:

• in_field – the name of the input label field

• out_field – the name of the output label field, which will be created if necessary

set_values(field_name, values, key_field=None, skip_none=False, expand_schema=True, dynamic=False, validate=True, progress=False, _allow_missing=False, _sample_ids=None, _frame_ids=None)

Sets the field or embedded field on each sample or frame in the collection to the given values.

You can use this method in two ways:

• Dict syntax (recommended): provide values as a dict whose keys specify the key_field values of the samples whose field_name you want to set to the corresponding values

• List syntax: provide values as a list, one for each sample in the collection on which you are invoking this method

Note

The most performant strategy for setting large numbers of field values is to use the dict syntax with key_field="id" when setting sample fields and key_field="frames.id" when setting frame fields. All other syntaxes internally convert to these IDs before ultimately performing the updates.

When setting a sample field embedded.field.name via the list values syntax, this function is an efficient implementation of the following loop:

for sample, value in zip(sample_collection, values):
    sample.embedded.field.name = value
    sample.save()

When setting an embedded field that contains an array, say embedded.array.field.name, via the list values syntax, this function is an efficient implementation of the following loop:

for sample, array_values in zip(sample_collection, values):
    for doc, value in zip(sample.embedded.array, array_values):
        doc.field.name = value

    sample.save()

When setting a frame field frames.embedded.field.name via the list values syntax, this function is an efficient implementation of the following loop:

for sample, frame_values in zip(sample_collection, values):
    for frame, value in zip(sample.frames.values(), frame_values):
        frame.embedded.field.name = value

    sample.save()

When setting an embedded frame field that contains an array, say frames.embedded.array.field.name, via the list values syntax, this function is an efficient implementation of the following loop:

for sample, frame_values in zip(sample_collection, values):
    for frame, array_values in zip(sample.frames.values(), frame_values):
        for doc, value in zip(frame.embedded.array, array_values):
            doc.field.name = value

    sample.save()

When setting a sample field embedded.field.name via the dict values syntax, this function is an efficient implementation of the following loop:

for key, value in values.items():
    sample = sample_collection.one(F(key_field) == key)
    sample.embedded.field.name = value
    sample.save()

When setting frame fields using the dict values syntax with a frame-level key_field, this function is an efficient implementation of the following loop:

frames = sample_collection.to_frames(...)
for key, value in values.items():
    frame = frames.one(F(key_field) == key)
    frame.embedded.field.name = value
    frame.save()

When setting frame fields using the dict values syntax with a sample-level key_field, each value in values may either be a list corresponding to the frames of the sample matching the given key, or each value may itself be a dict mapping frame numbers to values. In the latter case, this function is an efficient implementation of the following loop:

for key, frame_values in values.items():
    sample = sample_collection.one(F(key_field) == key)
    for frame_number, value in frame_values.items():
        frame = sample[frame_number]
        frame.embedded.field.name = value

    sample.save()

You can also update list fields using the dict values syntaxes, in which case this method is an efficient implementation of the natural nested list modifications of the above sample/frame loops.

The dual function of set_values() is values(), which can be used to efficiently extract the values of a field or embedded field of all samples in a collection as lists of values.

Note

If you are setting attributes of a nested list of labels, such as attributes of the objects in a fiftyone.core.labels.Detections field, then consider using set_label_values() instead for greater efficiency.

Note

If the values you are setting can be described by a fiftyone.core.expressions.ViewExpression applied to the existing dataset contents, then consider using set_field() + save() for an even more efficient alternative to explicitly iterating over the dataset or calling values() + set_values() to perform the update in-memory.

Examples:

import random

import fiftyone as fo
import fiftyone.zoo as foz
from fiftyone import ViewField as F

dataset = foz.load_zoo_dataset("quickstart")

#
# Create a new sample field
#

# list syntax
values = [random.random() for _ in range(len(dataset))]

dataset.set_values("random", values)

print(dataset.bounds("random"))

#
# Edit a frame field
#

# dict syntax
sample_ids = dataset.values("id")
values = {id: random.random() for id in sample_ids}

dataset.set_values("random", values, key_field="id")

print(dataset.bounds("random"))

#
# Add a tag to all low confidence labels
#

view = dataset.filter_labels("predictions", F("confidence") < 0.06)

# list syntax on a filtered view
tags = view.values("predictions.detections.tags")
for sample_tags in tags:
    for detection_tags in sample_tags:
        detection_tags.append("low_confidence")

view.set_values("predictions.detections.tags", tags)

print(view.count("predictions.detections"))  # 447
print(dataset.count_label_tags())  # 447

#
# Create a new frame field
#

dataset = foz.load_zoo_dataset("quickstart-video")

# list syntax
values = []
for sample in dataset:
    values.append([random.random() for _ in sample.frames])

dataset.set_values("frames.random", values)

print(dataset.bounds("frames.random"))

#
# Edit a frame field
#

# dict syntax
frame_ids = dataset.values("frames.id", unwind=True)
values = {id: random.random() for id in frame_ids}

dataset.set_values("frames.random", values, key_field="frames.id")

print(dataset.bounds("frames.random"))

Parameters:

• field_name – a field or embedded.field.name

• values –

  the field values to set, provided in either of the following formats:

  • list syntax: an iterable of values, one for each sample in the collection. If field_name contains array fields, the corresponding elements of values must be arrays of the same lengths. When setting frame fields, each element can either be an iterable of values (one for each existing frame of the sample) or a dict mapping frame numbers to values

  • dict syntax: a dict whose keys specify the key_field values of the samples for which to set field_name to the corresponding values. When setting frame fields, you can either provide a sample-level key_field, in which case each corresponding value in values must be a list or dict of per-frame field values to set as described in the previous bullet, or you can provide a frame-level key_field, in which case each key-value pair in values represents a per-frame update

• key_field (None) – a key field to use when choosing which samples to update when values is a dict

• skip_none (False) – whether to treat None data in values as missing data that should not be set

• expand_schema (True) – whether to dynamically add new sample/frame fields encountered to the dataset schema. If False, an error is raised if the root field_name does not exist

• dynamic (False) – whether to declare dynamic attributes of embedded document fields that are encountered

• validate (True) – whether to validate that the values are compliant with the dataset schema before adding them

• progress (False) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead

set_label_values(field_name, values, dynamic=False, skip_none=False, validate=True, progress=False)

Sets the fields of the specified labels in the collection to the given values.

You can use this method in two ways:

• List syntax (recommended): provide a list of dicts of the form {"sample_id": sample_id, "label_id": label_id, "value": value} specifying the sample IDs and label IDs of each label you want to edit

• Dict syntax: provide a dict mapping label IDs to values

Note

This method is most efficient when you use the list syntax, which includes the sample/frame ID of each label that you are modifying.

Note

This method is appropriate when you have the IDs of the labels you wish to modify. See set_values() and set_field() if your updates are not keyed by label ID.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz
from fiftyone import ViewField as F

dataset = foz.load_zoo_dataset("quickstart")

#
# Populate a new boolean attribute on all high confidence labels
#

view = dataset.filter_labels("predictions", F("confidence") > 0.99)

# Option 1 (recommended): provide label IDs and sample IDs

values = []
sample_ids, label_ids = view.values(["id", "predictions.detections.id"])
for sid, lids in zip(sample_ids, label_ids):
    for lid in lids:
        values.append({"sample_id": sid, "label_id": lid, "value": True})

dataset.set_label_values("predictions.detections.high_conf", values)

print(dataset.count("predictions.detections"))
print(len(values))
print(dataset.count_values("predictions.detections.high_conf"))

# Option 2: provide only label IDs

label_ids = view.values("predictions.detections.id", unwind=True)
values = {_id: True for _id in label_ids}

dataset.set_label_values("predictions.detections.high_conf", values)

print(dataset.count("predictions.detections"))
print(len(label_ids))
print(dataset.count_values("predictions.detections.high_conf"))

Parameters:

• field_name – a field or embedded.field.name

• values –

  the label values to set, in one of the following formats:

  • a list of dicts of the form {"sample_id": sample_id, "label_id": label_id, "value": value} when setting sample-level labels

  • a list of dicts of the form {"frame_id": frame_id, "label_id": label_id, "value": value} when setting frame-level labels

  • a dict mapping label IDs to values

• skip_none (False) – whether to treat None data in values as missing data that should not be set

• dynamic (False) – whether to declare dynamic attributes of embedded document fields that are encountered

• validate (True) – whether to validate that the values are compliant with the dataset schema before adding them

• progress (False) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead

compute_metadata(overwrite=False, num_workers=None, skip_failures=True, warn_failures=False, progress=None)

Populates the metadata field of all samples in the collection.

Any samples with existing metadata are skipped, unless overwrite == True.

Parameters:

• overwrite (False) – whether to overwrite existing metadata

• num_workers (None) – a suggested number of threads to use

• skip_failures (True) – whether to gracefully continue without raising an error if metadata cannot be computed for a sample

• warn_failures (False) – whether to log a warning if metadata cannot be computed for a sample

• progress (None) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead

apply_model(model, label_field='predictions', confidence_thresh=None, store_logits=False, batch_size=None, num_workers=None, skip_failures=True, output_dir=None, rel_dir=None, progress=None, **kwargs)

Applies the model to the samples in the collection.

This method supports all of the following cases:

• Applying an image model to an image collection

• Applying an image model to the frames of a video collection

• Applying a video model to a video collection

Parameters:

• model – a fiftyone.core.models.Model, Hugging Face transformers model, Ultralytics model, SuperGradients model, or Lightning Flash model

• label_field ("predictions") – the name of the field in which to store the model predictions. When performing inference on video frames, the “frames.” prefix is optional

• confidence_thresh (None) – an optional confidence threshold to apply to any applicable labels generated by the model

• store_logits (False) – whether to store logits for the model predictions. This is only supported when the provided model has logits, model.has_logits == True

• batch_size (None) – an optional batch size to use, if the model supports batching

• num_workers (None) – the number of workers for the torch.utils.data.DataLoader to use. Only applicable for Torch-based models

• skip_failures (True) – whether to gracefully continue without raising an error if predictions cannot be generated for a sample. Only applicable to fiftyone.core.models.Model instances

• output_dir (None) – an optional output directory in which to write segmentation images. Only applicable if the model generates segmentations. If none is provided, the segmentations are stored in the database

• rel_dir (None) – an optional relative directory to strip from each input filepath to generate a unique identifier that is joined with output_dir to generate an output path for each segmentation image. This argument allows for populating nested subdirectories in output_dir that match the shape of the input paths. The path is converted to an absolute path (if necessary) via fiftyone.core.storage.normalize_path()

• progress (None) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead

• **kwargs – optional model-specific keyword arguments passed through to the underlying inference implementation

compute_embeddings(model, embeddings_field=None, batch_size=None, num_workers=None, skip_failures=True, progress=None, **kwargs)

Computes embeddings for the samples in the collection using the given model.

This method supports all the following cases:

• Using an image model to compute embeddings for an image collection

• Using an image model to compute frame embeddings for a video collection

• Using a video model to compute embeddings for a video collection

The model must expose embeddings, i.e., fiftyone.core.models.Model.has_embeddings() must return True.

If an embeddings_field is provided, the embeddings are saved to the samples; otherwise, the embeddings are returned in-memory.

Parameters:

• model – a fiftyone.core.models.Model, Hugging Face Transformers model, Ultralytics model, SuperGradients model, or Lightning Flash model

• embeddings_field (None) – the name of a field in which to store the embeddings. When computing video frame embeddings, the “frames.” prefix is optional

• batch_size (None) – an optional batch size to use, if the model supports batching

• num_workers (None) – the number of workers for the torch.utils.data.DataLoader to use. Only applicable for Torch-based models

• skip_failures (True) – whether to gracefully continue without raising an error if embeddings cannot be generated for a sample. Only applicable to fiftyone.core.models.Model instances

• progress (None) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead

• **kwargs – optional model-specific keyword arguments passed through to the underlying inference implementation

Returns:

• None, if an embeddings_field is provided

• a num_samples x num_dim array of embeddings, when computing embeddings for image/video collections with image/video models, respectively, and no embeddings_field is provided. If skip_failures is True and any errors are detected, a list of length num_samples is returned instead containing all successfully computed embedding vectors along with None entries for samples for which embeddings could not be computed

• a dictionary mapping sample IDs to num_frames x num_dim arrays of embeddings, when computing frame embeddings for video collections using an image model. If skip_failures is True and any errors are detected, the values of this dictionary will contain arrays of embeddings for all frames 1, 2, … until the error occurred, or None if no embeddings were computed at all

Return type:

one of the following

compute_patch_embeddings(model, patches_field, embeddings_field=None, force_square=False, alpha=None, handle_missing='skip', batch_size=None, num_workers=None, skip_failures=True, progress=None)

Computes embeddings for the image patches defined by patches_field of the samples in the collection using the given model.

This method supports all the following cases:

• Using an image model to compute patch embeddings for an image collection

• Using an image model to compute frame patch embeddings for a video collection

The model must expose embeddings, i.e., fiftyone.core.models.Model.has_embeddings() must return True.

If an embeddings_field is provided, the embeddings are saved to the samples; otherwise, the embeddings are returned in-memory.

Parameters:

• model – a fiftyone.core.models.Model, Hugging Face Transformers model, Ultralytics model, SuperGradients model, or Lightning Flash model

• patches_field – the name of the field defining the image patches in each sample to embed. Must be of type fiftyone.core.labels.Detection, fiftyone.core.labels.Detections, fiftyone.core.labels.Polyline, or fiftyone.core.labels.Polylines. When computing video frame embeddings, the “frames.” prefix is optional

• embeddings_field (None) – the name of a label attribute in which to store the embeddings

• force_square (False) – whether to minimally manipulate the patch bounding boxes into squares prior to extraction

• alpha (None) – an optional expansion/contraction to apply to the patches before extracting them, in [-1, inf). If provided, the length and width of the box are expanded (or contracted, when alpha < 0) by (100 * alpha)%. For example, set alpha = 0.1 to expand the boxes by 10%, and set alpha = -0.1 to contract the boxes by 10%

• handle_missing ("skip") –

  how to handle images with no patches. Supported values are:

  • ”skip”: skip the image and assign its embedding as None

  • ”image”: use the whole image as a single patch

  • ”error”: raise an error

• batch_size (None) – an optional batch size to use, if the model supports batching

• num_workers (None) – the number of workers for the torch.utils.data.DataLoader to use. Only applicable for Torch-based models

• skip_failures (True) – whether to gracefully continue without raising an error if embeddings cannot be generated for a sample

• progress (None) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead

Returns:

• None, if an embeddings_field is provided

• a dict mapping sample IDs to num_patches x num_dim arrays of patch embeddings, when computing patch embeddings for image collections and no embeddings_field is provided. If skip_failures is True and any errors are detected, this dictionary will contain None values for any samples for which embeddings could not be computed

• a dict of dicts mapping sample IDs to frame numbers to num_patches x num_dim arrays of patch embeddings, when computing patch embeddings for the frames of video collections and no embeddings_field is provided. If skip_failures is True and any errors are detected, this nested dict will contain missing or None values to indicate uncomputable embeddings

Return type:

one of the following

evaluate_regressions(pred_field, gt_field='ground_truth', eval_key=None, missing=None, method=None, progress=None, **kwargs)

Evaluates the regression predictions in this collection with respect to the specified ground truth values.

You can customize the evaluation method by passing additional parameters for the method’s config class as kwargs.

The natively provided method values and their associated configs are:

• "simple": fiftyone.utils.eval.regression.SimpleEvaluationConfig

If an eval_key is specified, then this method will record some statistics on each sample:

• When evaluating sample-level fields, an eval_key field will be populated on each sample recording the error of that sample’s prediction.

• When evaluating frame-level fields, an eval_key field will be populated on each frame recording the error of that frame’s prediction. In addition, an eval_key field will be populated on each sample that records the average error of the frame predictions of the sample.

Parameters:

• pred_field – the name of the field containing the predicted fiftyone.core.labels.Regression instances

• gt_field ("ground_truth") – the name of the field containing the ground truth fiftyone.core.labels.Regression instances

• eval_key (None) – a string key to use to refer to this evaluation

• missing (None) – a missing value. Any None-valued regressions are given this value for results purposes

• method (None) – a string specifying the evaluation method to use. The supported values are fo.evaluation_config.regression_backends.keys() and the default is fo.evaluation_config.regression_default_backend

• progress (None) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead

• **kwargs – optional keyword arguments for the constructor of the fiftyone.utils.eval.regression.RegressionEvaluationConfig being used

Returns:

a fiftyone.utils.eval.regression.RegressionResults

evaluate_classifications(pred_field, gt_field='ground_truth', eval_key=None, classes=None, missing=None, method=None, progress=None, **kwargs)

Evaluates the classification predictions in this collection with respect to the specified ground truth labels.

By default, this method simply compares the ground truth and prediction for each sample, but other strategies such as binary evaluation and top-k matching can be configured via the method parameter.

You can customize the evaluation method by passing additional parameters for the method’s config class as kwargs.

The natively provided method values and their associated configs are:

• "simple": fiftyone.utils.eval.classification.SimpleEvaluationConfig

• "top-k": fiftyone.utils.eval.classification.TopKEvaluationConfig

• "binary": fiftyone.utils.eval.classification.BinaryEvaluationConfig

If an eval_key is specified, then this method will record some statistics on each sample:

• When evaluating sample-level fields, an eval_key field will be populated on each sample recording whether that sample’s prediction is correct.

• When evaluating frame-level fields, an eval_key field will be populated on each frame recording whether that frame’s prediction is correct. In addition, an eval_key field will be populated on each sample that records the average accuracy of the frame predictions of the sample.

Parameters:

• pred_field – the name of the field containing the predicted fiftyone.core.labels.Classification instances

• gt_field ("ground_truth") – the name of the field containing the ground truth fiftyone.core.labels.Classification instances

• eval_key (None) – a string key to use to refer to this evaluation

• classes (None) – the list of possible classes. If not provided, the observed ground truth/predicted labels are used

• missing (None) – a missing label string. Any None-valued labels are given this label for results purposes

• method (None) – a string specifying the evaluation method to use. The supported values are fo.evaluation_config.classification_backends.keys() and the default is fo.evaluation_config.classification_default_backend

• progress (None) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead

• **kwargs – optional keyword arguments for the constructor of the fiftyone.utils.eval.classification.ClassificationEvaluationConfig being used

Returns:

a fiftyone.utils.eval.classification.ClassificationResults

evaluate_detections(pred_field, gt_field='ground_truth', eval_key=None, classes=None, missing=None, method=None, iou=0.5, use_masks=False, use_boxes=False, classwise=True, dynamic=True, progress=None, **kwargs)

Evaluates the specified predicted detections in this collection with respect to the specified ground truth detections.

This method supports evaluating the following spatial data types:

• Object detections in fiftyone.core.labels.Detections format

• Instance segmentations in fiftyone.core.labels.Detections format with their mask attributes populated

• Polygons in fiftyone.core.labels.Polylines format

• Keypoints in fiftyone.core.labels.Keypoints format

• Temporal detections in fiftyone.core.labels.TemporalDetections format

For spatial object detection evaluation, this method uses COCO-style evaluation by default.

When evaluating keypoints, “IoUs” are computed via object keypoint similarity.

For temporal segment detection, this method uses ActivityNet-style evaluation by default.

You can use the method parameter to select a different method, and you can optionally customize the method by passing additional parameters for the method’s config class as kwargs.

The natively provided method values and their associated configs are:

• "coco": fiftyone.utils.eval.coco.COCOEvaluationConfig

• "open-images": fiftyone.utils.eval.openimages.OpenImagesEvaluationConfig

• "activitynet": fiftyone.utils.eval.activitynet.ActivityNetEvaluationConfig

If an eval_key is provided, a number of fields are populated at the object- and sample-level recording the results of the evaluation:

• True positive (TP), false positive (FP), and false negative (FN) counts for the each sample are saved in top-level fields of each sample:

  TP: sample.<eval_key>_tp
  FP: sample.<eval_key>_fp
  FN: sample.<eval_key>_fn
  

  In addition, when evaluating frame-level objects, TP/FP/FN counts are recorded for each frame:

  TP: frame.<eval_key>_tp
  FP: frame.<eval_key>_fp
  FN: frame.<eval_key>_fn
  

• The fields listed below are populated on each individual object; these fields tabulate the TP/FP/FN status of the object, the ID of the matching object (if any), and the matching IoU:

  TP/FP/FN: object.<eval_key>
        ID: object.<eval_key>_id
       IoU: object.<eval_key>_iou
  

Parameters:

• pred_field – the name of the field containing the predicted fiftyone.core.labels.Detections, fiftyone.core.labels.Polylines, fiftyone.core.labels.Keypoints, or fiftyone.core.labels.TemporalDetections

• gt_field ("ground_truth") – the name of the field containing the ground truth fiftyone.core.labels.Detections, fiftyone.core.labels.Polylines, fiftyone.core.labels.Keypoints, or fiftyone.core.labels.TemporalDetections

• eval_key (None) – a string key to use to refer to this evaluation

• classes (None) – the list of possible classes. If not provided, the observed ground truth/predicted labels are used

• missing (None) – a missing label string. Any unmatched objects are given this label for results purposes

• method (None) – a string specifying the evaluation method to use. The supported values are fo.evaluation_config.detection_backends.keys() and the default is fo.evaluation_config.detection_default_backend

• iou (0.50) – the IoU threshold to use to determine matches

• use_masks (False) – whether to compute IoUs using the instances masks in the mask attribute of the provided objects, which must be fiftyone.core.labels.Detection instances

• use_boxes (False) – whether to compute IoUs using the bounding boxes of the provided fiftyone.core.labels.Polyline instances rather than using their actual geometries

• classwise (True) – whether to only match objects with the same class label (True) or allow matches between classes (False)

• dynamic (True) – whether to declare the dynamic object-level attributes that are populated on the dataset’s schema

• progress (None) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead

• **kwargs – optional keyword arguments for the constructor of the fiftyone.utils.eval.detection.DetectionEvaluationConfig being used

Returns:

a fiftyone.utils.eval.detection.DetectionResults

evaluate_segmentations(pred_field, gt_field='ground_truth', eval_key=None, mask_targets=None, method=None, progress=None, **kwargs)

Evaluates the specified semantic segmentation masks in this collection with respect to the specified ground truth masks.

If the size of a predicted mask does not match the ground truth mask, it is resized to match the ground truth.

By default, this method simply performs pixelwise evaluation of the full masks, but other strategies such as boundary-only evaluation can be configured by passing additional parameters for the method’s config class as kwargs.

The natively provided method values and their associated configs are:

• "simple": fiftyone.utils.eval.segmentation.SimpleEvaluationConfig

If an eval_key is provided, the accuracy, precision, and recall of each sample is recorded in top-level fields of each sample:

 Accuracy: sample.<eval_key>_accuracy
Precision: sample.<eval_key>_precision
   Recall: sample.<eval_key>_recall

In addition, when evaluating frame-level masks, the accuracy, precision, and recall of each frame if recorded in the following frame-level fields:

 Accuracy: frame.<eval_key>_accuracy
Precision: frame.<eval_key>_precision
   Recall: frame.<eval_key>_recall

Note

The mask values 0 and #000000 are treated as a background class for the purposes of computing evaluation metrics like precision and recall.

Parameters:

• pred_field – the name of the field containing the predicted fiftyone.core.labels.Segmentation instances

• gt_field ("ground_truth") – the name of the field containing the ground truth fiftyone.core.labels.Segmentation instances

• eval_key (None) – a string key to use to refer to this evaluation

• mask_targets (None) – a dict mapping pixel values or RGB hex strings to labels. If not provided, the observed values are used as labels

• method (None) – a string specifying the evaluation method to use. The supported values are fo.evaluation_config.segmentation_backends.keys() and the default is fo.evaluation_config.segmentation_default_backend

• progress (None) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead

• **kwargs – optional keyword arguments for the constructor of the fiftyone.utils.eval.segmentation.SegmentationEvaluationConfig being used

Returns:

a fiftyone.utils.eval.segmentation.SegmentationResults

property has_evaluations

Whether this collection has any evaluation results.

has_evaluation(eval_key)

Whether this collection has an evaluation with the given key.

Parameters:

eval_key – an evaluation key

Returns:

True/False

list_evaluations(type=None, method=None, **kwargs)

Returns a list of evaluation keys on this collection.

Parameters:

• type (None) –

  a specific evaluation type to match, which can be:

  • a string fiftyone.core.evaluations.EvaluationMethodConfig.type

  • a fiftyone.core.evaluations.EvaluationMethod class or its fully-qualified class name string

• method (None) – a specific fiftyone.core.evaluations.EvaluationMethodConfig.method string to match

• **kwargs – optional config parameters to match

Returns:

a list of evaluation keys

map_samples(map_fcn, save=False, skip_failures=False, parallelize_method=None, num_workers=None, batch_method=None, batch_size=None, progress=None)

Applies the given function to each sample in the collection and returns the results as a generator.

By default, a multiprocessing pool is used to parallelize the work, unless this method is called in a daemon process (subprocess), in which case no workers are used.

This function effectively performs the following map operation with the outer loop in parallel:

for batch_view in fou.iter_slices(sample_collection, batch_size):
    for sample in batch_view.iter_samples(autosave=save):
        sample_output = map_fcn(sample)
        yield sample.id, sample_output

Example:

from collections import Counter

import fiftyone as fo
import fiftyone.zoo as foz

dataset = foz.load_zoo_dataset("cifar10", split="train")
view = dataset.select_fields("ground_truth")

def map_fcn(sample):
    return sample.ground_truth.label.upper()

counter = Counter()
for _, label in view.map_samples(map_fcn):
    counter[label] += 1

print(dict(counter))

Parameters:

• map_fcn – a function to apply to each sample in the collection

• save (False) – whether to save any sample edits applied by map_fcn

• skip_failures (False) – whether to gracefully continue without raising an error if the update function raises an exception for a sample

• parallelize_method (None) – the parallelization method to use. Supported values are {"process", "thread"}. The default is fiftyone.config.default_parallelization_method

• num_workers (None) – the number of workers to use. When using process parallelism, this defaults to fiftyone.config.default_process_pool_workers if the value is set, else fiftyone.core.utils.recommend_process_pool_workers() workers are used. If this value is <= 1, all work is done in the main process

• batch_method (None) – whether to use IDs ("id") or slices ("slice") to assign samples to workers

• batch_size (None) – an optional number of samples to distribute to each worker at a time. By default, samples are evenly distributed to workers with one batch per worker

• progress (None) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead, or “workers” to render per-worker progress bars

Returns:

a generator that emits (sample_id, map_output) tuples

update_samples(update_fcn, skip_failures=True, parallelize_method=None, num_workers=None, batch_method=None, batch_size=None, progress=None)

Applies the given function to each sample in the collection and saves the resulting sample edits.

By default, a multiprocessing pool is used to parallelize the work, unless this method is called in a daemon process (subprocess), in which case no workers are used.

This function effectively performs the following map operation with the outer loop in parallel:

for batch_view in fou.iter_slices(sample_collection, batch_size):
    for sample in batch_view.iter_samples(autosave=True):
        map_fcn(sample)

Example:

import fiftyone as fo
import fiftyone.zoo as foz

dataset = foz.load_zoo_dataset("cifar10", split="train")
view = dataset.select_fields("ground_truth")

def update_fcn(sample):
    sample.ground_truth.label = sample.ground_truth.label.upper()

view.update_samples(update_fcn)

print(dataset.count_values("ground_truth.label"))

Parameters:

• update_fcn – a function to apply to each sample in the collection

• skip_failures (True) – whether to gracefully continue without raising an error if the update function raises an exception for a sample

• parallelize_method (None) – the parallelization method to use. Supported values are {"process", "thread"}. The default is fiftyone.config.default_parallelization_method

• num_workers (None) – the number of workers to use. When using process parallelism, this defaults to fiftyone.config.default_process_pool_workers if the value is set, else fiftyone.core.utils.recommend_process_pool_workers() workers are used. If this value is <= 1, all work is done in the main process

• batch_method (None) – whether to use IDs ("id") or slices ("slice") to assign samples to workers

• batch_size (None) – an optional number of samples to distribute to each worker at a time. By default, samples are evenly distributed to workers with one batch per worker

• progress (None) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead, or “workers” to render per-worker progress bars

rename_evaluation(eval_key, new_eval_key)

Replaces the key for the given evaluation with a new key.

Parameters:

• eval_key – an evaluation key

• new_anno_key – a new evaluation key

get_evaluation_info(eval_key)

Returns information about the evaluation with the given key on this collection.

Parameters:

eval_key – an evaluation key

Returns:

an fiftyone.core.evaluation.EvaluationInfo

load_evaluation_results(eval_key, cache=True, **kwargs)

Loads the results for the evaluation with the given key on this collection.

Parameters:

• eval_key – an evaluation key

• cache (True) – whether to cache the results on the collection

• **kwargs – keyword arguments for the run’s fiftyone.core.evaluation.EvaluationMethodConfig.load_credentials() method

Returns:

a fiftyone.core.evaluation.EvaluationResults

load_evaluation_view(eval_key, select_fields=False)

Loads the fiftyone.core.view.DatasetView on which the specified evaluation was performed on this collection.

Parameters:

• eval_key – an evaluation key

• select_fields (False) – whether to exclude fields involved in other evaluations

Returns:

a fiftyone.core.view.DatasetView

delete_evaluation(eval_key)

Deletes the evaluation results associated with the given evaluation key from this collection.

Parameters:

eval_key – an evaluation key

delete_evaluations()

Deletes all evaluation results from this collection.

property has_brain_runs

Whether this collection has any brain runs.

has_brain_run(brain_key)

Whether this collection has a brain method run with the given key.

Parameters:

brain_key – a brain key

Returns:

True/False

list_brain_runs(type=None, method=None, **kwargs)

Returns a list of brain keys on this collection.

Parameters:

• type (None) –

  a specific brain run type to match, which can be:

  • a string fiftyone.core.brain.BrainMethodConfig.type

  • a fiftyone.core.brain.BrainMethod class or its fully-qualified class name string

• method (None) – a specific fiftyone.core.brain.BrainMethodConfig.method string to match

• **kwargs – optional config parameters to match

Returns:

a list of brain keys

rename_brain_run(brain_key, new_brain_key)

Replaces the key for the given brain run with a new key.

Parameters:

• brain_key – a brain key

• new_brain_key – a new brain key

get_brain_info(brain_key)

Returns information about the brain method run with the given key on this collection.

Parameters:

brain_key – a brain key

Returns:

a fiftyone.core.brain.BrainInfo

load_brain_results(brain_key, cache=True, load_view=True, **kwargs)

Loads the results for the brain method run with the given key on this collection.

Parameters:

• brain_key – a brain key

• cache (True) – whether to cache the results on the collection

• load_view (True) – whether to load the view on which the results were computed (True) or the full dataset (False)

• **kwargs – keyword arguments for the run’s fiftyone.core.brain.BrainMethodConfig.load_credentials() method

Returns:

a fiftyone.core.brain.BrainResults

load_brain_view(brain_key, select_fields=False)

Loads the fiftyone.core.view.DatasetView on which the specified brain method run was performed on this collection.

Parameters:

• brain_key – a brain key

• select_fields (False) – whether to exclude fields involved in other brain method runs

Returns:

a fiftyone.core.view.DatasetView

delete_brain_run(brain_key)

Deletes the brain method run with the given key from this collection.

Parameters:

brain_key – a brain key

delete_brain_runs()

Deletes all brain method runs from this collection.

property has_runs

Whether this collection has any runs.

has_run(run_key)

Whether this collection has a run with the given key.

Parameters:

run_key – a run key

Returns:

True/False

list_runs(**kwargs)

Returns a list of run keys on this collection.

Parameters:

**kwargs – optional config parameters to match

Returns:

a list of run keys

init_run(**kwargs)

Initializes a config instance for a new run.

Parameters:

**kwargs – JSON serializable config parameters

Returns:

a fiftyone.core.runs.RunConfig

register_run(run_key, config, results=None, overwrite=False, cleanup=True, cache=True)

Registers a run under the given key on this collection.

Parameters:

• run_key – a run key

• config – a fiftyone.core.runs.RunConfig

• results (None) – an optional fiftyone.core.runs.RunResults

• overwrite (False) – whether to allow overwriting an existing run of the same type

• cleanup (True) – whether to execute an existing run’s fiftyone.core.runs.Run.cleanup() method when overwriting it

• cache (True) – whether to cache the results on the collection

rename_run(run_key, new_run_key)

Replaces the key for the given run with a new key.

Parameters:

• run_key – a run key

• new_run_key – a new run key

get_run_info(run_key)

Returns information about the run with the given key on this collection.

Parameters:

run_key – a run key

Returns:

a fiftyone.core.runs.RunInfo

update_run_config(run_key, config)

Updates the run config for the run with the given key.

Parameters:

• run_key – a run key

• config – a fiftyone.core.runs.RunConfig

init_run_results(run_key, **kwargs)

Initializes a results instance for the run with the given key.

Parameters:

• run_key – a run key

• **kwargs – JSON serializable data

Returns:

a fiftyone.core.runs.RunResults

save_run_results(run_key, results, overwrite=True, cache=True)

Saves run results for the run with the given key.

Parameters:

• run_key – a run key

• results – a fiftyone.core.runs.RunResults

• overwrite (True) – whether to overwrite an existing result with the same key

• cache (True) – whether to cache the results on the collection

load_run_results(run_key, cache=True, load_view=True, **kwargs)

Loads the results for the run with the given key on this collection.

Parameters:

• run_key – a run key

• cache (True) – whether to cache the results on the collection

• load_view (True) – whether to load the view on which the results were computed (True) or the full dataset (False)

• **kwargs – keyword arguments for the run’s fiftyone.core.runs.RunConfig.load_credentials() method

Returns:

a fiftyone.core.runs.RunResults

load_run_view(run_key, select_fields=False)

Loads the fiftyone.core.view.DatasetView on which the specified run was performed on this collection.

Parameters:

• run_key – a run key

• select_fields (False) – whether to exclude fields involved in other runs

Returns:

a fiftyone.core.view.DatasetView

delete_run(run_key)

Deletes the run with the given key from this collection.

Parameters:

run_key – a run key

delete_runs()

Deletes all runs from this collection.

classmethod list_view_stages()

Returns a list of all available methods on this collection that apply fiftyone.core.stages.ViewStage operations to this collection.

Returns:

a list of SampleCollection method names

add_stage(stage)

Applies the given fiftyone.core.stages.ViewStage to the collection.

Parameters:

stage – a fiftyone.core.stages.ViewStage

Returns:

a fiftyone.core.view.DatasetView

concat(samples)

Concatenates the contents of the given SampleCollection to this collection.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz
from fiftyone import ViewField as F

dataset = foz.load_zoo_dataset("quickstart")

#
# Concatenate two views
#

view1 = dataset.match(F("uniqueness") < 0.2)
view2 = dataset.match(F("uniqueness") > 0.7)

view = view1.concat(view2)

print(view1)
print(view2)
print(view)

#
# Concatenate two patches views
#

gt_objects = dataset.to_patches("ground_truth")

patches1 = gt_objects[:50]
patches2 = gt_objects[-50:]
patches = patches1.concat(patches2)

print(patches1)
print(patches2)
print(patches)

Parameters:

samples – a SampleCollection whose contents to append to this collection

Returns:

a fiftyone.core.view.DatasetView

exclude(sample_ids)

Excludes the samples with the given IDs from the collection.

Examples:

import fiftyone as fo

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(filepath="/path/to/image1.png"),
        fo.Sample(filepath="/path/to/image2.png"),
        fo.Sample(filepath="/path/to/image3.png"),
    ]
)

#
# Exclude the first sample from the dataset
#

sample_id = dataset.first().id
view = dataset.exclude(sample_id)

#
# Exclude the first and last samples from the dataset
#

sample_ids = [dataset.first().id, dataset.last().id]
view = dataset.exclude(sample_ids)

Parameters:

sample_ids –

the samples to exclude. Can be any of the following:

• a sample ID

• an iterable of sample IDs

• a fiftyone.core.sample.Sample or fiftyone.core.sample.SampleView

• an iterable of fiftyone.core.sample.Sample or fiftyone.core.sample.SampleView instances

• a fiftyone.core.collections.SampleCollection

Returns:

a fiftyone.core.view.DatasetView

exclude_by(field, values)

Excludes the samples with the given field values from the collection.

This stage is typically used to work with categorical fields (strings, ints, and bools). If you want to exclude samples based on floating point fields, use match().

Examples:

import fiftyone as fo

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(filepath="image%d.jpg" % i, int=i, str=str(i))
        for i in range(10)
    ]
)

#
# Create a view excluding samples whose `int` field have the given
# values
#

view = dataset.exclude_by("int", [1, 9, 3, 7, 5])
print(view.head(5))

#
# Create a view excluding samples whose `str` field have the given
# values
#

view = dataset.exclude_by("str", ["1", "9", "3", "7", "5"])
print(view.head(5))

Parameters:

• field – a field or embedded.field.name

• values – a value or iterable of values to exclude by

Returns:

a fiftyone.core.view.DatasetView

exclude_fields(field_names=None, meta_filter=None, _allow_missing=False)

Excludes the fields with the given names from the samples in the collection.

Note that default fields cannot be excluded.

Examples:

import fiftyone as fo

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            ground_truth=fo.Classification(label="cat"),
            predictions=fo.Classification(
                label="cat",
                confidence=0.9,
                mood="surly",
            ),
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            ground_truth=fo.Classification(label="dog"),
            predictions=fo.Classification(
                label="dog",
                confidence=0.8,
                mood="happy",
            ),
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
        ),
    ]
)

#
# Exclude the `predictions` field from all samples
#

view = dataset.exclude_fields("predictions")

#
# Exclude the `mood` attribute from all classifications in the
# `predictions` field
#

view = dataset.exclude_fields("predictions.mood")

Parameters:

• field_names (None) – a field name or iterable of field names to exclude. May contain embedded.field.name as well

• meta_filter (None) –

  a filter that dynamically excludes fields in the collection’s schema according to the specified rule, which can be matched against the field’s name, type, description, and/or info. For example:

  • Use meta_filter="2023" or meta_filter={"any": "2023"} to exclude fields that have the string “2023” anywhere in their name, type, description, or info

  • Use meta_filter={"type": "StringField"} or meta_filter={"type": "Classification"} to exclude all string or classification fields, respectively

  • Use meta_filter={"description": "my description"} to exclude fields whose description contains the string “my description”

  • Use meta_filter={"info": "2023"} to exclude fields that have the string “2023” anywhere in their info

  • Use meta_filter={"info.key": "value"}} to exclude fields that have a specific key/value pair in their info

  • Include meta_filter={"include_nested_fields": True, ...} in your meta filter to include all nested fields in the filter

Returns:

a fiftyone.core.view.DatasetView

exclude_frames(frame_ids, omit_empty=True)

Excludes the frames with the given IDs from the video collection.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz

dataset = foz.load_zoo_dataset("quickstart-video")

#
# Exclude some specific frames
#

frame_ids = [
    dataset.first().frames.first().id,
    dataset.last().frames.last().id,
]

view = dataset.exclude_frames(frame_ids)

print(dataset.count("frames"))
print(view.count("frames"))

Parameters:

• frame_ids –

  the frames to exclude. Can be any of the following:

  • a frame ID

  • an iterable of frame IDs

  • a fiftyone.core.frame.Frame or fiftyone.core.frame.FrameView

  • an iterable of fiftyone.core.frame.Frame or fiftyone.core.frame.FrameView instances

  • a fiftyone.core.collections.SampleCollection whose frames to exclude

• omit_empty (True) – whether to omit samples that have no frames after excluding the specified frames

Returns:

a fiftyone.core.view.DatasetView

exclude_groups(group_ids)

Excludes the groups with the given IDs from the grouped collection.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz

dataset = foz.load_zoo_dataset("quickstart-groups")

#
# Exclude some specific groups by ID
#

view = dataset.take(2)
group_ids = view.values("group.id")
other_groups = dataset.exclude_groups(group_ids)

assert len(set(group_ids) & set(other_groups.values("group.id"))) == 0

Parameters:

groups_ids –

the groups to exclude. Can be any of the following:

• a group ID

• an iterable of group IDs

• a fiftyone.core.sample.Sample or fiftyone.core.sample.SampleView

• a group dict returned by get_group()

• an iterable of fiftyone.core.sample.Sample or fiftyone.core.sample.SampleView instances

• an iterable of group dicts returned by get_group()

• a fiftyone.core.collections.SampleCollection

Returns:

a fiftyone.core.view.DatasetView

exclude_group_slices(slices=None, media_type=None)

Excludes the specified group slice(s) from the grouped collection.

Examples:

import fiftyone as fo

dataset = fo.Dataset()
dataset.add_group_field("group", default="ego")

group1 = fo.Group()
group2 = fo.Group()

dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/left-image1.jpg",
            group=group1.element("left"),
        ),
        fo.Sample(
            filepath="/path/to/video1.mp4",
            group=group1.element("ego"),
        ),
        fo.Sample(
            filepath="/path/to/right-image1.jpg",
            group=group1.element("right"),
        ),
        fo.Sample(
            filepath="/path/to/left-image2.jpg",
            group=group2.element("left"),
        ),
        fo.Sample(
            filepath="/path/to/video2.mp4",
            group=group2.element("ego"),
        ),
        fo.Sample(
            filepath="/path/to/right-image2.jpg",
            group=group2.element("right"),
        ),
    ]
)

#
# Exclude the samples from the "ego" group slice
#

view = dataset.exclude_group_slices("ego")

#
# Exclude the samples from the "left" or "right" group slices
#

view = dataset.exclude_group_slices(["left", "right"])

#
# Exclude all image slices
#

view = dataset.exclude_group_slices(media_type="image")

Parameters:

• slices (None) – a group slice or iterable of group slices to exclude

• media_type (None) – a media type or iterable of media types whose slice(s) to exclude

Returns:

a fiftyone.core.view.DatasetView

exclude_labels(labels=None, ids=None, instance_ids=None, tags=None, fields=None, omit_empty=True)

Excludes the specified labels from the collection.

The returned view will omit samples, sample fields, and individual labels that do not match the specified selection criteria.

You can perform an exclusion via one or more of the following methods:

• Provide the labels argument, which should contain a list of dicts in the format returned by fiftyone.core.session.Session.selected_labels, to exclude specific labels

• Provide the ids argument to exclude labels with specific IDs

• Provide the instance_ids argument to exclude labels with specific instance IDs

• Provide the tags argument to exclude labels with specific tags

If multiple criteria are specified, labels must match all of them in order to be excluded.

By default, the exclusion is applied to all fiftyone.core.labels.Label fields, but you can provide the fields argument to explicitly define the field(s) in which to exclude.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz

dataset = foz.load_zoo_dataset("quickstart")

#
# Exclude the labels currently selected in the App
#

session = fo.launch_app(dataset)

# Select some labels in the App...

view = dataset.exclude_labels(labels=session.selected_labels)

#
# Exclude labels with the specified IDs
#

# Grab some label IDs
ids = [
    dataset.first().ground_truth.detections[0].id,
    dataset.last().predictions.detections[0].id,
]

view = dataset.exclude_labels(ids=ids)

print(dataset.count("ground_truth.detections"))
print(view.count("ground_truth.detections"))

print(dataset.count("predictions.detections"))
print(view.count("predictions.detections"))

#
# Exclude labels with the specified tags
#

# Grab some label IDs
ids = [
    dataset.first().ground_truth.detections[0].id,
    dataset.last().predictions.detections[0].id,
]

# Give the labels a "test" tag
dataset = dataset.clone()  # create copy since we're modifying data
dataset.select_labels(ids=ids).tag_labels("test")

print(dataset.count_values("ground_truth.detections.tags"))
print(dataset.count_values("predictions.detections.tags"))

# Exclude the labels via their tag
view = dataset.exclude_labels(tags="test")

print(dataset.count("ground_truth.detections"))
print(view.count("ground_truth.detections"))

print(dataset.count("predictions.detections"))
print(view.count("predictions.detections"))

Parameters:

• labels (None) – a list of dicts specifying the labels to exclude in the format returned by fiftyone.core.session.Session.selected_labels

• ids (None) – an ID or iterable of IDs of the labels to exclude

• instance_ids (None) – an instance ID or iterable of instance IDs of the labels to exclude

• tags (None) – a tag or iterable of tags of labels to exclude

• fields (None) – a field or iterable of fields from which to exclude

• omit_empty (True) – whether to omit samples that have no labels after filtering

Returns:

a fiftyone.core.view.DatasetView

exists(field, bool=None)

Returns a view containing the samples in the collection that have (or do not have) a non-None value for the given field or embedded field.

Examples:

import fiftyone as fo

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            ground_truth=fo.Classification(label="cat"),
            predictions=fo.Classification(label="cat", confidence=0.9),
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            ground_truth=fo.Classification(label="dog"),
            predictions=fo.Classification(label="dog", confidence=0.8),
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            ground_truth=fo.Classification(label="dog"),
            predictions=fo.Classification(label="dog"),
        ),
        fo.Sample(
            filepath="/path/to/image4.png",
            ground_truth=None,
            predictions=None,
        ),
        fo.Sample(filepath="/path/to/image5.png"),
    ]
)

#
# Only include samples that have a value in their `predictions`
# field
#

view = dataset.exists("predictions")

#
# Only include samples that do NOT have a value in their
# `predictions` field
#

view = dataset.exists("predictions", False)

#
# Only include samples that have prediction confidences
#

view = dataset.exists("predictions.confidence")

Parameters:

• field – the field name or embedded.field.name

• bool (None) – whether to check if the field exists (None or True) or does not exist (False)

Returns:

a fiftyone.core.view.DatasetView

filter_field(field, filter, only_matches=True)

Filters the values of a field or embedded field of each sample in the collection.

Values of field for which filter returns False are replaced with None.

Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            ground_truth=fo.Classification(label="cat"),
            predictions=fo.Classification(label="cat", confidence=0.9),
            numeric_field=1.0,
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            ground_truth=fo.Classification(label="dog"),
            predictions=fo.Classification(label="dog", confidence=0.8),
            numeric_field=-1.0,
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            ground_truth=None,
            predictions=None,
            numeric_field=None,
        ),
    ]
)

#
# Only include classifications in the `predictions` field
# whose `label` is "cat"
#

view = dataset.filter_field("predictions", F("label") == "cat")

#
# Only include samples whose `numeric_field` value is positive
#

view = dataset.filter_field("numeric_field", F() > 0)

Parameters:

• field – the field name or embedded.field.name

• filter –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression that returns a boolean describing the filter to apply

• only_matches (True) – whether to only include samples that match the filter (True) or include all samples (False)

Returns:

a fiftyone.core.view.DatasetView

filter_labels(field, filter, only_matches=True, trajectories=False)

Filters the fiftyone.core.labels.Label field of each sample in the collection.

If the specified field is a single fiftyone.core.labels.Label type, fields for which filter returns False are replaced with None:

• fiftyone.core.labels.Classification

• fiftyone.core.labels.Detection

• fiftyone.core.labels.Polyline

• fiftyone.core.labels.Keypoint

If the specified field is a fiftyone.core.labels.Label list type, the label elements for which filter returns False are omitted from the view:

• fiftyone.core.labels.Classifications

• fiftyone.core.labels.Detections

• fiftyone.core.labels.Polylines

• fiftyone.core.labels.Keypoints

Classifications Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            predictions=fo.Classification(label="cat", confidence=0.9),
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            predictions=fo.Classification(label="dog", confidence=0.8),
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            predictions=fo.Classification(label="rabbit"),
        ),
        fo.Sample(
            filepath="/path/to/image4.png",
            predictions=None,
        ),
    ]
)

#
# Only include classifications in the `predictions` field whose
# `confidence` is greater than 0.8
#

view = dataset.filter_labels("predictions", F("confidence") > 0.8)

#
# Only include classifications in the `predictions` field whose
# `label` is "cat" or "dog"
#

view = dataset.filter_labels(
    "predictions", F("label").is_in(["cat", "dog"])
)

Detections Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            predictions=fo.Detections(
                detections=[
                    fo.Detection(
                        label="cat",
                        bounding_box=[0.1, 0.1, 0.5, 0.5],
                        confidence=0.9,
                    ),
                    fo.Detection(
                        label="dog",
                        bounding_box=[0.2, 0.2, 0.3, 0.3],
                        confidence=0.8,
                    ),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            predictions=fo.Detections(
                detections=[
                    fo.Detection(
                        label="cat",
                        bounding_box=[0.5, 0.5, 0.4, 0.4],
                        confidence=0.95,
                    ),
                    fo.Detection(label="rabbit"),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            predictions=fo.Detections(
                detections=[
                    fo.Detection(
                        label="squirrel",
                        bounding_box=[0.25, 0.25, 0.5, 0.5],
                        confidence=0.5,
                    ),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image4.png",
            predictions=None,
        ),
    ]
)

#
# Only include detections in the `predictions` field whose
# `confidence` is greater than 0.8
#

view = dataset.filter_labels("predictions", F("confidence") > 0.8)

#
# Only include detections in the `predictions` field whose `label`
# is "cat" or "dog"
#

view = dataset.filter_labels(
    "predictions", F("label").is_in(["cat", "dog"])
)

#
# Only include detections in the `predictions` field whose bounding
# box area is smaller than 0.2
#

# Bboxes are in [top-left-x, top-left-y, width, height] format
bbox_area = F("bounding_box")[2] * F("bounding_box")[3]

view = dataset.filter_labels("predictions", bbox_area < 0.2)

Polylines Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            predictions=fo.Polylines(
                polylines=[
                    fo.Polyline(
                        label="lane",
                        points=[[(0.1, 0.1), (0.1, 0.6)]],
                        filled=False,
                    ),
                    fo.Polyline(
                        label="road",
                        points=[[(0.2, 0.2), (0.5, 0.5), (0.2, 0.5)]],
                        filled=True,
                    ),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            predictions=fo.Polylines(
                polylines=[
                    fo.Polyline(
                        label="lane",
                        points=[[(0.4, 0.4), (0.9, 0.4)]],
                        filled=False,
                    ),
                    fo.Polyline(
                        label="road",
                        points=[[(0.6, 0.6), (0.9, 0.9), (0.6, 0.9)]],
                        filled=True,
                    ),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            predictions=None,
        ),
    ]
)

#
# Only include polylines in the `predictions` field that are filled
#

view = dataset.filter_labels("predictions", F("filled") == True)

#
# Only include polylines in the `predictions` field whose `label`
# is "lane"
#

view = dataset.filter_labels("predictions", F("label") == "lane")

#
# Only include polylines in the `predictions` field with at least
# 3 vertices
#

num_vertices = F("points").map(F().length()).sum()
view = dataset.filter_labels("predictions", num_vertices >= 3)

Keypoints Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            predictions=fo.Keypoint(
                label="house",
                points=[(0.1, 0.1), (0.1, 0.9), (0.9, 0.9), (0.9, 0.1)],
            ),
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            predictions=fo.Keypoint(
                label="window",
                points=[(0.4, 0.4), (0.5, 0.5), (0.6, 0.6)],
            ),
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            predictions=None,
        ),
    ]
)

#
# Only include keypoints in the `predictions` field whose `label`
# is "house"
#

view = dataset.filter_labels("predictions", F("label") == "house")

#
# Only include keypoints in the `predictions` field with less than
# four points
#

view = dataset.filter_labels("predictions", F("points").length() < 4)

Parameters:

• field – the label field to filter

• filter –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression that returns a boolean describing the filter to apply

• only_matches (True) – whether to only include samples with at least one label after filtering (True) or include all samples (False)

• trajectories (False) – whether to match entire object trajectories for which the object matches the given filter on at least one frame. Only applicable to datasets that contain videos and frame-level label fields whose objects have their index attributes populated

Returns:

a fiftyone.core.view.DatasetView

filter_keypoints(field, filter=None, labels=None, only_matches=True)

Filters the individual fiftyone.core.labels.Keypoint.points elements in the specified keypoints field of each sample in the collection.

Note

Use filter_labels() if you simply want to filter entire fiftyone.core.labels.Keypoint objects in a field.

Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            predictions=fo.Keypoints(
                keypoints=[
                    fo.Keypoint(
                        label="person",
                        points=[(0.1, 0.1), (0.1, 0.9), (0.9, 0.9), (0.9, 0.1)],
                        confidence=[0.7, 0.8, 0.95, 0.99],
                    )
                ]
            )
        ),
        fo.Sample(filepath="/path/to/image2.png"),
    ]
)

dataset.default_skeleton = fo.KeypointSkeleton(
    labels=["nose", "left eye", "right eye", "left ear", "right ear"],
    edges=[[0, 1, 2, 0], [0, 3], [0, 4]],
)

#
# Only include keypoints in the `predictions` field whose
# `confidence` is greater than 0.9
#

view = dataset.filter_keypoints(
    "predictions", filter=F("confidence") > 0.9
)

#
# Only include keypoints in the `predictions` field with less than
# four points
#

view = dataset.filter_keypoints(
    "predictions", labels=["left eye", "right eye"]
)

Parameters:

• field – the fiftyone.core.labels.Keypoint or fiftyone.core.labels.Keypoints field to filter

• filter (None) –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression that returns a boolean, like F("confidence") > 0.5 or F("occluded") == False, to apply elementwise to the specified field, which must be a list of same length as fiftyone.core.labels.Keypoint.points

• labels (None) – a label or iterable of keypoint skeleton labels to keep

• only_matches (True) – whether to only include keypoints/samples with at least one point after filtering (True) or include all keypoints/samples (False)

Returns:

a fiftyone.core.view.DatasetView

flatten(stages=None)

Returns a flattened view that contains all samples in the dynamic grouped collection.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz
from fiftyone import ViewField as F

dataset = foz.load_zoo_dataset("cifar10", split="test")

# Group samples by ground truth label
grouped_view = dataset.take(1000).group_by("ground_truth.label")
print(len(grouped_view))  # 10

# Return a flat view that contains 10 samples from each class
flat_view = grouped_view.flatten(fo.Limit(10))
print(len(flat_view))  # 100

Parameters:

stages (None) – a fiftyone.core.stages.ViewStage or list of fiftyone.core.stages.ViewStage instances to apply to each group’s samples while flattening

Returns:

a fiftyone.core.view.DatasetView

geo_near(point, location_field=None, min_distance=None, max_distance=None, query=None, create_index=True)

Sorts the samples in the collection by their proximity to a specified geolocation.

Note

This stage must be the first stage in any fiftyone.core.view.DatasetView in which it appears.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz

TIMES_SQUARE = [-73.9855, 40.7580]

dataset = foz.load_zoo_dataset("quickstart-geo")

#
# Sort the samples by their proximity to Times Square
#

view = dataset.geo_near(TIMES_SQUARE)

#
# Sort the samples by their proximity to Times Square, and only
# include samples within 5km
#

view = dataset.geo_near(TIMES_SQUARE, max_distance=5000)

#
# Sort the samples by their proximity to Times Square, and only
# include samples that are in Manhattan
#

import fiftyone.utils.geojson as foug

in_manhattan = foug.geo_within(
    "location.point",
    [
        [
            [-73.949701, 40.834487],
            [-73.896611, 40.815076],
            [-73.998083, 40.696534],
            [-74.031751, 40.715273],
            [-73.949701, 40.834487],
        ]
    ]
)

view = dataset.geo_near(
    TIMES_SQUARE, location_field="location", query=in_manhattan
)

Parameters:

• point –

  the reference point to compute distances to. Can be any of the following:

  • A [longitude, latitude] list

  • A GeoJSON dict with Point type

  • A fiftyone.core.labels.GeoLocation instance whose point attribute contains the point

• location_field (None) –

  the location data of each sample to use. Can be any of the following:

  • The name of a fiftyone.core.fields.GeoLocation field whose point attribute to use as location data

  • An embedded.field.name containing GeoJSON data to use as location data

  • None, in which case there must be a single fiftyone.core.fields.GeoLocation field on the samples, which is used by default

• min_distance (None) – filter samples that are less than this distance (in meters) from point

• max_distance (None) – filter samples that are greater than this distance (in meters) from point

• query (None) – an optional dict defining a MongoDB read query that samples must match in order to be included in this view

• create_index (True) – whether to create the required spherical index, if necessary

Returns:

a fiftyone.core.view.DatasetView

geo_within(boundary, location_field=None, strict=True, create_index=True)

Filters the samples in this collection to only include samples whose geolocation is within a specified boundary.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz

MANHATTAN = [
    [
        [-73.949701, 40.834487],
        [-73.896611, 40.815076],
        [-73.998083, 40.696534],
        [-74.031751, 40.715273],
        [-73.949701, 40.834487],
    ]
]

dataset = foz.load_zoo_dataset("quickstart-geo")

#
# Create a view that only contains samples in Manhattan
#

view = dataset.geo_within(MANHATTAN)

Parameters:

• boundary – a fiftyone.core.labels.GeoLocation, fiftyone.core.labels.GeoLocations, GeoJSON dict, or list of coordinates that define a Polygon or MultiPolygon to search within

• location_field (None) –

  the location data of each sample to use. Can be any of the following:

  • The name of a fiftyone.core.fields.GeoLocation field whose point attribute to use as location data

  • An embedded.field.name that directly contains the GeoJSON location data to use

  • None, in which case there must be a single fiftyone.core.fields.GeoLocation field on the samples, which is used by default

• strict (True) – whether a sample’s location data must strictly fall within boundary (True) in order to match, or whether any intersection suffices (False)

• create_index (True) – whether to create the required spherical index, if necessary

Returns:

a fiftyone.core.view.DatasetView

group_by(field_or_expr, order_by=None, reverse=False, flat=False, match_expr=None, sort_expr=None, create_index=True, order_by_key=None)

Creates a view that groups the samples in the collection by a specified field or expression.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz
from fiftyone import ViewField as F

dataset = foz.load_zoo_dataset("cifar10", split="test")

#
# Take 1000 samples at random and group them by ground truth label
#

view = dataset.take(1000).group_by("ground_truth.label")

for group in view.iter_dynamic_groups():
    group_value = group.first().ground_truth.label
    print("%s: %d" % (group_value, len(group)))

#
# Variation of above operation that arranges the groups in
# decreasing order of size and immediately flattens them
#

from itertools import groupby

view = dataset.take(1000).group_by(
    "ground_truth.label",
    flat=True,
    sort_expr=F().length(),
    reverse=True,
)

rle = lambda v: [(k, len(list(g))) for k, g in groupby(v)]
for label, count in rle(view.values("ground_truth.label")):
    print("%s: %d" % (label, count))

Parameters:

• field_or_expr – the field or embedded.field.name to group by, or a list of field names defining a compound group key, or a fiftyone.core.expressions.ViewExpression or MongoDB aggregation expression that defines the value to group by

• order_by (None) – an optional field by which to order the samples in each group

• reverse (False) – whether to return the results in descending order Applies both to order_by and sort_expr

• flat (False) – whether to return a grouped collection (False) or a flattened collection (True)

• match_expr (None) –

  an optional fiftyone.core.expressions.ViewExpression or MongoDB aggregation expression that defines which groups to include in the output view. If provided, this expression will be evaluated on the list of samples in each group. Only applicable when flat=True

• sort_expr (None) –

  an optional fiftyone.core.expressions.ViewExpression or MongoDB aggregation expression that defines how to sort the groups in the output view. If provided, this expression will be evaluated on the list of samples in each group. Only applicable when flat=True

• create_index (True) – whether to create an index, if necessary, to optimize the grouping. Only applicable when grouping by field(s), not expressions

• order_by_key (None) – an optional fixed order_by value representing the first sample in a group. Required for optimized performance. See this guide for more details

Returns:

a fiftyone.core.view.DatasetView

limit(limit)

Returns a view with at most the given number of samples.

Examples:

import fiftyone as fo

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            ground_truth=fo.Classification(label="cat"),
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            ground_truth=fo.Classification(label="dog"),
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            ground_truth=None,
        ),
    ]
)

#
# Only include the first 2 samples in the view
#

view = dataset.limit(2)

Parameters:

limit – the maximum number of samples to return. If a non-positive number is provided, an empty view is returned

Returns:

a fiftyone.core.view.DatasetView

limit_labels(field, limit)

Limits the number of fiftyone.core.labels.Label instances in the specified labels list field of each sample in the collection.

The specified field must be one of the following types:

• fiftyone.core.labels.Classifications

• fiftyone.core.labels.Detections

• fiftyone.core.labels.Keypoints

• fiftyone.core.labels.Polylines

Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            predictions=fo.Detections(
                detections=[
                    fo.Detection(
                        label="cat",
                        bounding_box=[0.1, 0.1, 0.5, 0.5],
                        confidence=0.9,
                    ),
                    fo.Detection(
                        label="dog",
                        bounding_box=[0.2, 0.2, 0.3, 0.3],
                        confidence=0.8,
                    ),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            predictions=fo.Detections(
                detections=[
                    fo.Detection(
                        label="cat",
                        bounding_box=[0.5, 0.5, 0.4, 0.4],
                        confidence=0.95,
                    ),
                    fo.Detection(label="rabbit"),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image4.png",
            predictions=None,
        ),
    ]
)

#
# Only include the first detection in the `predictions` field of
# each sample
#

view = dataset.limit_labels("predictions", 1)

Parameters:

• field – the labels list field to filter

• limit – the maximum number of labels to include in each labels list. If a non-positive number is provided, all lists will be empty

Returns:

a fiftyone.core.view.DatasetView

map_labels(field, map)

Maps the label values of a fiftyone.core.labels.Label field to new values for each sample in the collection.

Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            weather=fo.Classification(label="sunny"),
            predictions=fo.Detections(
                detections=[
                    fo.Detection(
                        label="cat",
                        bounding_box=[0.1, 0.1, 0.5, 0.5],
                        confidence=0.9,
                    ),
                    fo.Detection(
                        label="dog",
                        bounding_box=[0.2, 0.2, 0.3, 0.3],
                        confidence=0.8,
                    ),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            weather=fo.Classification(label="cloudy"),
            predictions=fo.Detections(
                detections=[
                    fo.Detection(
                        label="cat",
                        bounding_box=[0.5, 0.5, 0.4, 0.4],
                        confidence=0.95,
                    ),
                    fo.Detection(label="rabbit"),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            weather=fo.Classification(label="partly cloudy"),
            predictions=fo.Detections(
                detections=[
                    fo.Detection(
                        label="squirrel",
                        bounding_box=[0.25, 0.25, 0.5, 0.5],
                        confidence=0.5,
                    ),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image4.png",
            predictions=None,
        ),
    ]
)

#
# Map the "partly cloudy" weather label to "cloudy"
#

view = dataset.map_labels("weather", {"partly cloudy": "cloudy"})

#
# Map "rabbit" and "squirrel" predictions to "other"
#

view = dataset.map_labels(
    "predictions", {"rabbit": "other", "squirrel": "other"}
)

Parameters:

• field – the labels field to map

• map – a dict mapping label values to new label values

Returns:

a fiftyone.core.view.DatasetView

map_values(field, map)

Maps the values in the given field to new values for each sample in the collection.

Examples:

import random

import fiftyone as fo
import fiftyone.zoo as foz
from fiftyone import ViewField as F

ANIMALS = [
    "bear", "bird", "cat", "cow", "dog", "elephant", "giraffe",
    "horse", "sheep", "zebra"
]

dataset = foz.load_zoo_dataset("quickstart")

values = [random.choice(ANIMALS) for _ in range(len(dataset))]
dataset.set_values("str_field", values)
dataset.set_values("list_field", [[v] for v in values])

dataset.set_field("ground_truth.detections.tags", [F("label")]).save()

# Map all animals to string "animal"
mapping = {a: "animal" for a in ANIMALS}

#
# Map values in top-level fields
#

view = dataset.map_values("str_field", mapping)

print(view.count_values("str_field"))
# {"animal": 200}

view = dataset.map_values("list_field", mapping)

print(view.count_values("list_field"))
# {"animal": 200}

#
# Map values in nested fields
#

view = dataset.map_values("ground_truth.detections.label", mapping)

print(view.count_values("ground_truth.detections.label"))
# {"animal": 183, ...}

view = dataset.map_values("ground_truth.detections.tags", mapping)

print(view.count_values("ground_truth.detections.tags"))
# {"animal": 183, ...}

Parameters:

• field – the field or embedded.field.name to map

• map – a dict mapping values to new values

Returns:

a fiftyone.core.view.DatasetView

set_field(field, expr, _allow_missing=False)

Sets a field or embedded field on each sample in a collection by evaluating the given expression.

This method can process embedded list fields. To do so, simply append [] to any list component(s) of the field path.

Note

There are two cases where FiftyOne will automatically unwind array fields without requiring you to explicitly specify this via the [] syntax:

Top-level lists: when you specify a field path that refers to a top-level list field of a dataset; i.e., list_field is automatically coerced to list_field[], if necessary.

List fields: When you specify a field path that refers to the list field of a Label class, such as the Detections.detections attribute; i.e., ground_truth.detections.label is automatically coerced to ground_truth.detections[].label, if necessary.

See the examples below for demonstrations of this behavior.

The provided expr is interpreted relative to the document on which the embedded field is being set. For example, if you are setting a nested field field="embedded.document.field", then the expression expr you provide will be applied to the embedded.document document. Note that you can override this behavior by defining an expression that is bound to the root document by prepending "$" to any field name(s) in the expression.

See the examples below for more information.

Note

Note that you cannot set a non-existing top-level field using this stage, since doing so would violate the dataset’s schema. You can, however, first declare a new field via fiftyone.core.dataset.Dataset.add_sample_field() and then populate it in a view via this stage.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz
from fiftyone import ViewField as F

dataset = foz.load_zoo_dataset("quickstart")

#
# Replace all values of the `uniqueness` field that are less than
# 0.5 with `None`
#

view = dataset.set_field(
    "uniqueness",
    (F("uniqueness") >= 0.5).if_else(F("uniqueness"), None)
)
print(view.bounds("uniqueness"))

#
# Lower bound all object confidences in the `predictions` field at
# 0.5
#

view = dataset.set_field(
    "predictions.detections.confidence", F("confidence").max(0.5)
)
print(view.bounds("predictions.detections.confidence"))

#
# Add a `num_predictions` property to the `predictions` field that
# contains the number of objects in the field
#

view = dataset.set_field(
    "predictions.num_predictions",
    F("$predictions.detections").length(),
)
print(view.bounds("predictions.num_predictions"))

#
# Set an `is_animal` field on each object in the `predictions` field
# that indicates whether the object is an animal
#

ANIMALS = [
    "bear", "bird", "cat", "cow", "dog", "elephant", "giraffe",
    "horse", "sheep", "zebra"
]

view = dataset.set_field(
    "predictions.detections.is_animal", F("label").is_in(ANIMALS)
)
print(view.count_values("predictions.detections.is_animal"))

Parameters:

• field – the field or embedded.field.name to set

• expr –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression that defines the field value to set

Returns:

a fiftyone.core.view.DatasetView

match(filter)

Filters the samples in the collection by the given filter.

Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            weather=fo.Classification(label="sunny"),
            predictions=fo.Detections(
                detections=[
                    fo.Detection(
                        label="cat",
                        bounding_box=[0.1, 0.1, 0.5, 0.5],
                        confidence=0.9,
                    ),
                    fo.Detection(
                        label="dog",
                        bounding_box=[0.2, 0.2, 0.3, 0.3],
                        confidence=0.8,
                    ),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image2.jpg",
            weather=fo.Classification(label="cloudy"),
            predictions=fo.Detections(
                detections=[
                    fo.Detection(
                        label="cat",
                        bounding_box=[0.5, 0.5, 0.4, 0.4],
                        confidence=0.95,
                    ),
                    fo.Detection(label="rabbit"),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            weather=fo.Classification(label="partly cloudy"),
            predictions=fo.Detections(
                detections=[
                    fo.Detection(
                        label="squirrel",
                        bounding_box=[0.25, 0.25, 0.5, 0.5],
                        confidence=0.5,
                    ),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image4.jpg",
            predictions=None,
        ),
    ]
)

#
# Only include samples whose `filepath` ends with ".jpg"
#

view = dataset.match(F("filepath").ends_with(".jpg"))

#
# Only include samples whose `weather` field is "sunny"
#

view = dataset.match(F("weather").label == "sunny")

#
# Only include samples with at least 2 objects in their
# `predictions` field
#

view = dataset.match(F("predictions").detections.length() >= 2)

#
# Only include samples whose `predictions` field contains at least
# one object with area smaller than 0.2
#

# Bboxes are in [top-left-x, top-left-y, width, height] format
bbox = F("bounding_box")
bbox_area = bbox[2] * bbox[3]

small_boxes = F("predictions.detections").filter(bbox_area < 0.2)
view = dataset.match(small_boxes.length() > 0)

Parameters:

filter –

a fiftyone.core.expressions.ViewExpression or MongoDB expression that returns a boolean describing the filter to apply

Returns:

a fiftyone.core.view.DatasetView

match_frames(filter, omit_empty=True)

Filters the frames in the video collection by the given filter.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz
from fiftyone import ViewField as F

dataset = foz.load_zoo_dataset("quickstart-video")

#
# Match frames with at least 10 detections
#

num_objects = F("detections.detections").length()
view = dataset.match_frames(num_objects > 10)

print(dataset.count())
print(view.count())

print(dataset.count("frames"))
print(view.count("frames"))

Parameters:

• filter –

  a fiftyone.core.expressions.ViewExpression or MongoDB aggregation expression that returns a boolean describing the filter to apply

• omit_empty (True) – whether to omit samples with no frame labels after filtering

Returns:

a fiftyone.core.view.DatasetView

match_labels(labels=None, ids=None, instance_ids=None, tags=None, filter=None, fields=None, bool=None)

Selects the samples from the collection that contain (or do not contain) at least one label that matches the specified criteria.

Note that, unlike select_labels() and filter_labels(), this stage will not filter the labels themselves; it only selects the corresponding samples.

You can perform a selection via one or more of the following methods:

• Provide the labels argument, which should contain a list of dicts in the format returned by fiftyone.core.session.Session.selected_labels, to match specific labels

• Provide the ids argument to match labels with specific IDs

• Provide the instance_ids argument to match labels with specific instance IDs

• Provide the tags argument to match labels with specific tags

• Provide the filter argument to match labels based on a boolean fiftyone.core.expressions.ViewExpression that is applied to each individual fiftyone.core.labels.Label element

• Pass bool=False to negate the operation and instead match samples that do not contain at least one label matching the specified criteria

If multiple criteria are specified, labels must match all of them in order to trigger a sample match.

By default, the selection is applied to all fiftyone.core.labels.Label fields, but you can provide the fields argument to explicitly define the field(s) in which to search.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz
from fiftyone import ViewField as F

dataset = foz.load_zoo_dataset("quickstart")

#
# Only show samples whose labels are currently selected in the App
#

session = fo.launch_app(dataset)

# Select some labels in the App...

view = dataset.match_labels(labels=session.selected_labels)

#
# Only include samples that contain labels with the specified IDs
#

# Grab some label IDs
ids = [
    dataset.first().ground_truth.detections[0].id,
    dataset.last().predictions.detections[0].id,
]

view = dataset.match_labels(ids=ids)

print(len(view))
print(view.count("ground_truth.detections"))
print(view.count("predictions.detections"))

#
# Only include samples that contain labels with the specified tags
#

# Grab some label IDs
ids = [
    dataset.first().ground_truth.detections[0].id,
    dataset.last().predictions.detections[0].id,
]

# Give the labels a "test" tag
dataset = dataset.clone()  # create copy since we're modifying data
dataset.select_labels(ids=ids).tag_labels("test")

print(dataset.count_values("ground_truth.detections.tags"))
print(dataset.count_values("predictions.detections.tags"))

# Retrieve the labels via their tag
view = dataset.match_labels(tags="test")

print(len(view))
print(view.count("ground_truth.detections"))
print(view.count("predictions.detections"))

#
# Only include samples that contain labels matching a filter
#

filter = F("confidence") > 0.99
view = dataset.match_labels(filter=filter, fields="predictions")

print(len(view))
print(view.count("ground_truth.detections"))
print(view.count("predictions.detections"))

Parameters:

• labels (None) – a list of dicts specifying the labels to select in the format returned by fiftyone.core.session.Session.selected_labels

• ids (None) – an ID or iterable of IDs of the labels to select

• instance_ids (None) – an instance ID or iterable of instance IDs of the labels to select

• tags (None) – a tag or iterable of tags of labels to select

• filter (None) –

  a fiftyone.core.expressions.ViewExpression or MongoDB aggregation expression that returns a boolean describing whether to select a given label. In the case of list fields like fiftyone.core.labels.Detections, the filter is applied to the list elements, not the root field

• fields (None) – a field or iterable of fields from which to select

• bool (None) – whether to match samples that have (None or True) or do not have (False) at least one label that matches the specified criteria

Returns:

a fiftyone.core.view.DatasetView

match_tags(tags, bool=None, all=False)

Returns a view containing the samples in the collection that have or don’t have any/all of the given tag(s).

Examples:

import fiftyone as fo

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(filepath="image1.png", tags=["train"]),
        fo.Sample(filepath="image2.png", tags=["test"]),
        fo.Sample(filepath="image3.png", tags=["train", "test"]),
        fo.Sample(filepath="image4.png"),
    ]
)

#
# Only include samples that have the "test" tag
#

view = dataset.match_tags("test")

#
# Only include samples that do not have the "test" tag
#

view = dataset.match_tags("test", bool=False)

#
# Only include samples that have the "test" or "train" tags
#

view = dataset.match_tags(["test", "train"])

#
# Only include samples that have the "test" and "train" tags
#

view = dataset.match_tags(["test", "train"], all=True)

#
# Only include samples that do not have the "test" or "train" tags
#

view = dataset.match_tags(["test", "train"], bool=False)

#
# Only include samples that do not have the "test" and "train" tags
#

view = dataset.match_tags(["test", "train"], bool=False, all=True)

Parameters:

• tags – the tag or iterable of tags to match

• bool (None) – whether to match samples that have (None or True) or do not have (False) the given tags

• all (False) – whether to match samples that have (or don’t have) all (True) or any (False) of the given tags

Returns:

a fiftyone.core.view.DatasetView

mongo(pipeline, _needs_frames=None, _group_slices=None)

Adds a view stage defined by a raw MongoDB aggregation pipeline.

See MongoDB aggregation pipelines for more details.

Examples:

import fiftyone as fo

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            predictions=fo.Detections(
                detections=[
                    fo.Detection(
                        label="cat",
                        bounding_box=[0.1, 0.1, 0.5, 0.5],
                        confidence=0.9,
                    ),
                    fo.Detection(
                        label="dog",
                        bounding_box=[0.2, 0.2, 0.3, 0.3],
                        confidence=0.8,
                    ),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            predictions=fo.Detections(
                detections=[
                    fo.Detection(
                        label="cat",
                        bounding_box=[0.5, 0.5, 0.4, 0.4],
                        confidence=0.95,
                    ),
                    fo.Detection(label="rabbit"),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            predictions=fo.Detections(
                detections=[
                    fo.Detection(
                        label="squirrel",
                        bounding_box=[0.25, 0.25, 0.5, 0.5],
                        confidence=0.5,
                    ),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image4.png",
            predictions=None,
        ),
    ]
)

#
# Extract a view containing the second and third samples in the
# dataset
#

view = dataset.mongo([{"$skip": 1}, {"$limit": 2}])

#
# Sort by the number of objects in the `precictions` field
#

view = dataset.mongo([
    {
        "$addFields": {
            "_sort_field": {
                "$size": {"$ifNull": ["$predictions.detections", []]}
            }
        }
    },
    {"$sort": {"_sort_field": -1}},
    {"$project": {"_sort_field": False}},
])

Parameters:

pipeline – a MongoDB aggregation pipeline (list of dicts)

Returns:

a fiftyone.core.view.DatasetView

select(sample_ids, ordered=False)

Selects the samples with the given IDs from the collection.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz

dataset = foz.load_zoo_dataset("quickstart")

#
# Create a view containing the currently selected samples in the App
#

session = fo.launch_app(dataset)

# Select samples in the App...

view = dataset.select(session.selected)

Parameters:

sample_ids –

the samples to select. Can be any of the following:

• a sample ID

• an iterable of sample IDs

• an iterable of booleans of same length as the collection encoding which samples to select

• a fiftyone.core.sample.Sample or fiftyone.core.sample.SampleView

• an iterable of fiftyone.core.sample.Sample or fiftyone.core.sample.SampleView instances

• a fiftyone.core.collections.SampleCollection

ordered (False): whether to sort the samples in the returned view to

match the order of the provided IDs

Returns:

a fiftyone.core.view.DatasetView

select_by(field, values, ordered=False)

Selects the samples with the given field values from the collection.

This stage is typically used to work with categorical fields (strings, ints, and bools). If you want to select samples based on floating point fields, use match().

Examples:

import fiftyone as fo

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(filepath="image%d.jpg" % i, int=i, str=str(i))
        for i in range(100)
    ]
)

#
# Create a view containing samples whose `int` field have the given
# values
#

view = dataset.select_by("int", [1, 51, 11, 41, 21, 31])
print(view.head(6))

#
# Create a view containing samples whose `str` field have the given
# values, in order
#

view = dataset.select_by(
    "str", ["1", "51", "11", "41", "21", "31"], ordered=True
)
print(view.head(6))

Parameters:

• field – a field or embedded.field.name

• values – a value or iterable of values to select by

• ordered (False) – whether to sort the samples in the returned view to match the order of the provided values

Returns:

a fiftyone.core.view.DatasetView

select_fields(field_names=None, meta_filter=None, _allow_missing=False)

Selects only the fields with the given names from the samples in the collection. All other fields are excluded.

Note that default sample fields are always selected.

Examples:

import fiftyone as fo

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            uniqueness=1.0,
            ground_truth=fo.Detections(
                detections=[
                    fo.Detection(
                        label="cat",
                        bounding_box=[0.1, 0.1, 0.5, 0.5],
                        mood="surly",
                        age=51,
                    ),
                    fo.Detection(
                        label="dog",
                        bounding_box=[0.2, 0.2, 0.3, 0.3],
                        mood="happy",
                        age=52,
                    ),
                ]
            )
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            uniqueness=0.0,
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
        ),
    ]
)

#
# Include only the default fields on each sample
#

view = dataset.select_fields()

#
# Include only the `uniqueness` field (and the default fields) on
# each sample
#

view = dataset.select_fields("uniqueness")

#
# Include only the `mood` attribute (and the default attributes) of
# each `Detection` in the `ground_truth` field
#

view = dataset.select_fields("ground_truth.detections.mood")

Parameters:

• field_names (None) – a field name or iterable of field names to select. May contain embedded.field.name as well

• meta_filter (None) –

  a filter that dynamically selects fields in the collection’s schema according to the specified rule, which can be matched against the field’s name, type, description, and/or info. For example:

  • Use meta_filter="2023" or meta_filter={"any": "2023"} to select fields that have the string “2023” anywhere in their name, type, description, or info

  • Use meta_filter={"type": "StringField"} or meta_filter={"type": "Classification"} to select all string or classification fields, respectively

  • Use meta_filter={"description": "my description"} to select fields whose description contains the string “my description”

  • Use meta_filter={"info": "2023"} to select fields that have the string “2023” anywhere in their info

  • Use meta_filter={"info.key": "value"}} to select fields that have a specific key/value pair in their info

  • Include meta_filter={"include_nested_fields": True, ...} in your meta filter to include all nested fields in the filter

Returns:

a fiftyone.core.view.DatasetView

select_frames(frame_ids, omit_empty=True)

Selects the frames with the given IDs from the video collection.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz

dataset = foz.load_zoo_dataset("quickstart-video")

#
# Select some specific frames
#

frame_ids = [
    dataset.first().frames.first().id,
    dataset.last().frames.last().id,
]

view = dataset.select_frames(frame_ids)

print(dataset.count())
print(view.count())

print(dataset.count("frames"))
print(view.count("frames"))

Parameters:

• frame_ids –

  the frames to select. Can be any of the following:

  • a frame ID

  • an iterable of frame IDs

  • a fiftyone.core.frame.Frame or fiftyone.core.frame.FrameView

  • an iterable of fiftyone.core.frame.Frame or fiftyone.core.frame.FrameView instances

  • a fiftyone.core.collections.SampleCollection whose frames to select

• omit_empty (True) – whether to omit samples that have no frames after selecting the specified frames

Returns:

a fiftyone.core.view.DatasetView

select_groups(group_ids, ordered=False)

Selects the groups with the given IDs from the grouped collection.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz

dataset = foz.load_zoo_dataset("quickstart-groups")

#
# Select some specific groups by ID
#

group_ids = dataset.take(10).values("group.id")

view = dataset.select_groups(group_ids)

assert set(view.values("group.id")) == set(group_ids)

view = dataset.select_groups(group_ids, ordered=True)

assert view.values("group.id") == group_ids

Parameters:

• groups_ids –

  the groups to select. Can be any of the following:

  • a group ID

  • an iterable of group IDs

  • a fiftyone.core.sample.Sample or fiftyone.core.sample.SampleView

  • a group dict returned by get_group()

  • an iterable of fiftyone.core.sample.Sample or fiftyone.core.sample.SampleView instances

  • an iterable of group dicts returned by get_group()

  • a fiftyone.core.collections.SampleCollection

• ordered (False) – whether to sort the groups in the returned view to match the order of the provided IDs

Returns:

a fiftyone.core.view.DatasetView

select_group_slices(slices=None, media_type=None, flat=True, _allow_mixed=False, _force_mixed=False)

Selects the specified group slice(s) from the grouped collection.

When flat==True, the returned view is a flattened non-grouped view containing the samples from the slice(s) of interest.

When flat=False, the returned view is a grouped collection containing only the slice(s) of interest.

Note

When flat=True, this stage performs a $lookup that pulls the requested slice(s) for each sample in the input collection from the source dataset. As a result, the stage emits unfiltered samples.

Examples:

import fiftyone as fo

dataset = fo.Dataset()
dataset.add_group_field("group", default="ego")

group1 = fo.Group()
group2 = fo.Group()

dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/left-image1.jpg",
            group=group1.element("left"),
        ),
        fo.Sample(
            filepath="/path/to/video1.mp4",
            group=group1.element("ego"),
        ),
        fo.Sample(
            filepath="/path/to/right-image1.jpg",
            group=group1.element("right"),
        ),
        fo.Sample(
            filepath="/path/to/left-image2.jpg",
            group=group2.element("left"),
        ),
        fo.Sample(
            filepath="/path/to/video2.mp4",
            group=group2.element("ego"),
        ),
        fo.Sample(
            filepath="/path/to/right-image2.jpg",
            group=group2.element("right"),
        ),
    ]
)

#
# Retrieve the samples from the "ego" group slice
#

view = dataset.select_group_slices("ego")

#
# Retrieve the samples from the "left" or "right" group slices
#

view = dataset.select_group_slices(["left", "right"])

#
# Select only the "left" and "right" group slices
#

view = dataset.select_group_slices(["left", "right"], flat=False)

#
# Retrieve all image samples
#

view = dataset.select_group_slices(media_type="image")

Parameters:

• slices (None) – a group slice or iterable of group slices to select. If neither argument is provided, a flattened list of all samples is returned

• media_type (None) – a media type or iterable of media types whose slice(s) to select

• flat (True) – whether to return a flattened collection (True) or a grouped collection (False)

Returns:

a fiftyone.core.view.DatasetView

select_labels(labels=None, ids=None, instance_ids=None, tags=None, fields=None, omit_empty=True)

Selects only the specified labels from the collection.

The returned view will omit samples, sample fields, and individual labels that do not match the specified selection criteria.

You can perform a selection via one or more of the following methods:

• Provide the labels argument, which should contain a list of dicts in the format returned by fiftyone.core.session.Session.selected_labels, to select specific labels

• Provide the ids argument to select labels with specific IDs

• Provide the instance_ids argument to select labels with specific instance IDs

• Provide the tags argument to select labels with specific tags

If multiple criteria are specified, labels must match all of them in order to be selected.

By default, the selection is applied to all fiftyone.core.labels.Label fields, but you can provide the fields argument to explicitly define the field(s) in which to select.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz

dataset = foz.load_zoo_dataset("quickstart")

#
# Only include the labels currently selected in the App
#

session = fo.launch_app(dataset)

# Select some labels in the App...

view = dataset.select_labels(labels=session.selected_labels)

#
# Only include labels with the specified IDs
#

# Grab some label IDs
ids = [
    dataset.first().ground_truth.detections[0].id,
    dataset.last().predictions.detections[0].id,
]

view = dataset.select_labels(ids=ids)

print(view.count("ground_truth.detections"))
print(view.count("predictions.detections"))

#
# Only include labels with the specified tags
#

# Grab some label IDs
ids = [
    dataset.first().ground_truth.detections[0].id,
    dataset.last().predictions.detections[0].id,
]

# Give the labels a "test" tag
dataset = dataset.clone()  # create copy since we're modifying data
dataset.select_labels(ids=ids).tag_labels("test")

print(dataset.count_label_tags())

# Retrieve the labels via their tag
view = dataset.select_labels(tags="test")

print(view.count("ground_truth.detections"))
print(view.count("predictions.detections"))

Parameters:

• labels (None) – a list of dicts specifying the labels to select in the format returned by fiftyone.core.session.Session.selected_labels

• ids (None) – an ID or iterable of IDs of the labels to select

• instance_ids (None) – an instance ID or iterable of instance IDs of the labels to select

• tags (None) – a tag or iterable of tags of labels to select

• fields (None) – a field or iterable of fields from which to select

• omit_empty (True) – whether to omit samples that have no labels after filtering

Returns:

a fiftyone.core.view.DatasetView

shuffle(seed=None)

Randomly shuffles the samples in the collection.

Examples:

import fiftyone as fo

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            ground_truth=fo.Classification(label="cat"),
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            ground_truth=fo.Classification(label="dog"),
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            ground_truth=None,
        ),
    ]
)

#
# Return a view that contains a randomly shuffled version of the
# samples in the dataset
#

view = dataset.shuffle()

#
# Shuffle the samples with a fixed random seed
#

view = dataset.shuffle(seed=51)

Parameters:

seed (None) – an optional random seed to use when shuffling the samples

Returns:

a fiftyone.core.view.DatasetView

skip(skip)

Omits the given number of samples from the head of the collection.

Examples:

import fiftyone as fo

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            ground_truth=fo.Classification(label="cat"),
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            ground_truth=fo.Classification(label="dog"),
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            ground_truth=fo.Classification(label="rabbit"),
        ),
        fo.Sample(
            filepath="/path/to/image4.png",
            ground_truth=None,
        ),
    ]
)

#
# Omit the first two samples from the dataset
#

view = dataset.skip(2)

Parameters:

skip – the number of samples to skip. If a non-positive number is provided, no samples are omitted

Returns:

a fiftyone.core.view.DatasetView

sort_by(field_or_expr, reverse=False, create_index=True)

Sorts the samples in the collection by the given field(s) or expression(s).

Examples:

import fiftyone as fo
import fiftyone.zoo as foz
from fiftyone import ViewField as F

dataset = foz.load_zoo_dataset("quickstart")

#
# Sort the samples by their `uniqueness` field in ascending order
#

view = dataset.sort_by("uniqueness", reverse=False)

#
# Sorts the samples in descending order by the number of detections
# in their `predictions` field whose bounding box area is less than
# 0.2
#

# Bboxes are in [top-left-x, top-left-y, width, height] format
bbox = F("bounding_box")
bbox_area = bbox[2] * bbox[3]

small_boxes = F("predictions.detections").filter(bbox_area < 0.2)
view = dataset.sort_by(small_boxes.length(), reverse=True)

#
# Performs a compound sort where samples are first sorted in
# descending or by number of detections and then in ascending order
# of uniqueness for samples with the same number of predictions
#

view = dataset.sort_by(
    [
        (F("predictions.detections").length(), -1),
        ("uniqueness", 1),
    ]
)

num_objects, uniqueness = view[:5].values(
    [F("predictions.detections").length(), "uniqueness"]
)
print(list(zip(num_objects, uniqueness)))

Parameters:

• field_or_expr –

  the field(s) or expression(s) to sort by. This can be any of the following:

  • a field to sort by

  • an embedded.field.name to sort by

  • a fiftyone.core.expressions.ViewExpression or a MongoDB aggregation expression that defines the quantity to sort by

  • a list of (field_or_expr, order) tuples defining a compound sort criteria, where field_or_expr is a field or expression as defined above, and order can be 1 or any string starting with “a” for ascending order, or -1 or any string starting with “d” for descending order

• reverse (False) – whether to return the results in descending order

• create_index (True) – whether to create an index, if necessary, to optimize the sort. Only applicable when sorting by field(s), not expressions

Returns:

a fiftyone.core.view.DatasetView

sort_by_similarity(query, k=None, reverse=False, dist_field=None, brain_key=None)

Sorts the collection by similarity to a specified query.

In order to use this stage, you must first use fiftyone.brain.compute_similarity() to index your dataset by similarity.

Examples:

import fiftyone as fo
import fiftyone.brain as fob
import fiftyone.zoo as foz

dataset = foz.load_zoo_dataset("quickstart")

fob.compute_similarity(
    dataset, model="clip-vit-base32-torch", brain_key="clip"
)

#
# Sort samples by their similarity to a sample by its ID
#

query_id = dataset.first().id

view = dataset.sort_by_similarity(query_id, k=5)

#
# Sort samples by their similarity to a manually computed vector
#

model = foz.load_zoo_model("clip-vit-base32-torch")
embeddings = dataset.take(2, seed=51).compute_embeddings(model)
query = embeddings.mean(axis=0)

view = dataset.sort_by_similarity(query, k=5)

#
# Sort samples by their similarity to a text prompt
#

query = "kites high in the air"

view = dataset.sort_by_similarity(query, k=5)

Parameters:

• query –

  the query, which can be any of the following:

  • an ID or iterable of IDs

  • a num_dims vector or num_queries x num_dims array of vectors

  • a prompt or iterable of prompts (if supported by the index)

• k (None) – the number of matches to return. By default, the entire collection is sorted

• reverse (False) – whether to sort by least similarity (True) or greatest similarity (False). Some backends may not support least similarity

• dist_field (None) – the name of a float field in which to store the distance of each example to the specified query. The field is created if necessary

• brain_key (None) – the brain key of an existing fiftyone.brain.compute_similarity() run on the dataset. If not specified, the dataset must have an applicable run, which will be used by default

Returns:

a fiftyone.core.view.DatasetView

take(size, seed=None)

Randomly samples the given number of samples from the collection.

Examples:

import fiftyone as fo

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            ground_truth=fo.Classification(label="cat"),
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            ground_truth=fo.Classification(label="dog"),
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            ground_truth=fo.Classification(label="rabbit"),
        ),
        fo.Sample(
            filepath="/path/to/image4.png",
            ground_truth=None,
        ),
    ]
)

#
# Take two random samples from the dataset
#

view = dataset.take(2)

#
# Take two random samples from the dataset with a fixed seed
#

view = dataset.take(2, seed=51)

Parameters:

• size – the number of samples to return. If a non-positive number is provided, an empty view is returned

• seed (None) – an optional random seed to use when selecting the samples

Returns:

a fiftyone.core.view.DatasetView

to_patches(field, **kwargs)

Creates a view that contains one sample per object patch in the specified field of the collection.

Fields other than field and the default sample fields will not be included in the returned view. A sample_id field will be added that records the sample ID from which each patch was taken.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz

dataset = foz.load_zoo_dataset("quickstart")

session = fo.launch_app(dataset)

#
# Create a view containing the ground truth patches
#

view = dataset.to_patches("ground_truth")
print(view)

session.view = view

Parameters:

• field – the patches field, which must be of type fiftyone.core.labels.Detections, fiftyone.core.labels.Polylines, or fiftyone.core.labels.Keypoints

• other_fields (None) –

  controls whether fields other than field and the default sample fields are included. Can be any of the following:

  • a field or list of fields to include

  • True to include all other fields

  • None/False to include no other fields

• keep_label_lists (False) – whether to store the patches in label list fields of the same type as the input collection rather than using their single label variants

• include_indexes (False) – whether to recreate any custom indexes on field and other_fields on the patches view (True) or a list of specific indexes or index prefixes to recreate. By default, no custom indexes are recreated

Returns:

a fiftyone.core.patches.PatchesView

to_evaluation_patches(eval_key, **kwargs)

Creates a view based on the results of the evaluation with the given key that contains one sample for each true positive, false positive, and false negative example in the collection, respectively.

True positive examples will result in samples with both their ground truth and predicted fields populated, while false positive/negative examples will only have one of their corresponding predicted/ground truth fields populated, respectively.

If multiple predictions are matched to a ground truth object (e.g., if the evaluation protocol includes a crowd attribute), then all matched predictions will be stored in the single sample along with the ground truth object.

The returned dataset will also have top-level type and iou fields populated based on the evaluation results for that example, as well as a sample_id field recording the sample ID of the example, and a crowd field if the evaluation protocol defines a crowd attribute.

Note

The returned view will contain patches for the contents of this collection, which may differ from the view on which the eval_key evaluation was performed. This may exclude some labels that were evaluated and/or include labels that were not evaluated.

If you would like to see patches for the exact view on which an evaluation was performed, first call load_evaluation_view() to load the view and then convert to patches.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz

dataset = foz.load_zoo_dataset("quickstart")
dataset.evaluate_detections("predictions", eval_key="eval")

session = fo.launch_app(dataset)

#
# Create a patches view for the evaluation results
#

view = dataset.to_evaluation_patches("eval")
print(view)

session.view = view

Parameters:

• eval_key – an evaluation key that corresponds to the evaluation of ground truth/predicted fields that are of type fiftyone.core.labels.Detections, fiftyone.core.labels.Polylines, or fiftyone.core.labels.Keypoints

• other_fields (None) –

  controls whether fields other than the ground truth/predicted fields and the default sample fields are included. Can be any of the following:

  • a field or list of fields to include

  • True to include all other fields

  • None/False to include no other fields

• include_indexes (False) – whether to recreate any custom indexes on the ground truth/predicted fields and other_fields on the patches view (True) or a list of specific indexes or index prefixes to recreate. By default, no custom indexes are recreated

Returns:

a fiftyone.core.patches.EvaluationPatchesView

to_clips(field_or_expr, **kwargs)

Creates a view that contains one sample per clip defined by the given field or expression in the video collection.

The returned view will contain:

• A sample_id field that records the sample ID from which each clip was taken

• A support field that records the [first, last] frame support of each clip

• All frame-level information from the underlying dataset of the input collection

Refer to fiftyone.core.clips.make_clips_dataset() to see the available configuration options for generating clips.

Note

The clip generation logic will respect any frame-level modifications defined in the input collection, but the output clips will always contain all frame-level labels.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz
from fiftyone import ViewField as F

dataset = foz.load_zoo_dataset("quickstart-video")

#
# Create a clips view that contains one clip for each contiguous
# segment that contains at least one road sign in every frame
#

clips = (
    dataset
    .filter_labels("frames.detections", F("label") == "road sign")
    .to_clips("frames.detections")
)
print(clips)

#
# Create a clips view that contains one clip for each contiguous
# segment that contains at least two road signs in every frame
#

signs = F("detections.detections").filter(F("label") == "road sign")
clips = dataset.to_clips(signs.length() >= 2)
print(clips)

Parameters:

• field_or_expr –

  can be any of the following:

  • a fiftyone.core.labels.TemporalDetection, fiftyone.core.labels.TemporalDetections, fiftyone.core.fields.FrameSupportField, or list of fiftyone.core.fields.FrameSupportField field

  • a frame-level label list field of any of the following types:

    • fiftyone.core.labels.Classifications

    • fiftyone.core.labels.Detections

    • fiftyone.core.labels.Polylines

    • fiftyone.core.labels.Keypoints

  • a fiftyone.core.expressions.ViewExpression that returns a boolean to apply to each frame of the input collection to determine if the frame should be clipped

  • a list of [(first1, last1), (first2, last2), ...] lists defining the frame numbers of the clips to extract from each sample

• other_fields (None) –

  controls whether sample fields other than the default sample fields are included. Can be any of the following:

  • a field or list of fields to include

  • True to include all other fields

  • None/False to include no other fields

• include_indexes (False) – whether to recreate any custom indexes on field_or_expr and other_fields on the clips view (True) or a list of specific indexes or index prefixes to recreate. By default, no custom indexes are recreated

• tol (0) – the maximum number of false frames that can be overlooked when generating clips. Only applicable when field_or_expr is a frame-level list field or expression

• min_len (0) – the minimum allowable length of a clip, in frames. Only applicable when field_or_expr is a frame-level list field or an expression

• trajectories (False) – whether to create clips for each unique object trajectory defined by their (label, index). Only applicable when field_or_expr is a frame-level field

Returns:

a fiftyone.core.clips.ClipsView

to_trajectories(field, **kwargs)

Creates a view that contains one clip for each unique object trajectory defined by their (label, index) in a frame-level field of a video collection.

The returned view will contain:

• A sample_id field that records the sample ID from which each clip was taken

• A support field that records the [first, last] frame support of each clip

• A sample-level label field that records the label and index of each trajectory

Examples:

import fiftyone as fo
import fiftyone.zoo as foz
from fiftyone import ViewField as F

dataset = foz.load_zoo_dataset("quickstart-video")

#
# Create a trajectories view for the vehicles in the dataset
#

trajectories = (
    dataset
    .filter_labels("frames.detections", F("label") == "vehicle")
    .to_trajectories("frames.detections")
)

print(trajectories)

Parameters:

• field –

  a frame-level label list field of any of the following types:

  • fiftyone.core.labels.Detections

  • fiftyone.core.labels.Polylines

  • fiftyone.core.labels.Keypoints

• other_fields (None) –

  controls whether sample fields other than the default sample fields are included. Can be any of the following:

  • a field or list of fields to include

  • True to include all other fields

  • None/False to include no other fields

• include_indexes (False) – whether to recreate any custom indexes on other_fields on the clips view (True) or a list of specific indexes or index prefixes to recreate. By default, no custom indexes are recreated

• tol (0) – the maximum number of false frames that can be overlooked when generating clips

• min_len (0) – the minimum allowable length of a clip, in frames

Returns:

a fiftyone.core.clips.TrajectoriesView

to_frames(**kwargs)

Creates a view that contains one sample per frame in the video collection.

The returned view will contain all frame-level fields and the tags of each video as sample-level fields, as well as a sample_id field that records the IDs of the parent sample for each frame.

By default, sample_frames is False and this method assumes that the frames of the input collection have filepath fields populated pointing to each frame image. Any frames without a filepath populated will be omitted from the returned view.

When sample_frames is True, this method samples each video in the collection into a directory of per-frame images and stores the filepaths in the filepath frame field of the source dataset. By default, each folder of images is written using the same basename as the input video. For example, if frames_patt = "%%06d.jpg", then videos with the following paths:

/path/to/video1.mp4
/path/to/video2.mp4
...

would be sampled as follows:

/path/to/video1/
    000001.jpg
    000002.jpg
    ...
/path/to/video2/
    000001.jpg
    000002.jpg
    ...

However, you can use the optional output_dir and rel_dir parameters to customize the location and shape of the sampled frame folders. For example, if output_dir = "/tmp" and rel_dir = "/path/to", then videos with the following paths:

/path/to/folderA/video1.mp4
/path/to/folderA/video2.mp4
/path/to/folderB/video3.mp4
...

would be sampled as follows:

/tmp/folderA/
    video1/
        000001.jpg
        000002.jpg
        ...
    video2/
        000001.jpg
        000002.jpg
        ...
/tmp/folderB/
    video3/
        000001.jpg
        000002.jpg
        ...

By default, samples will be generated for every video frame at full resolution, but this method provides a variety of parameters that can be used to customize the sampling behavior.

Note

If this method is run multiple times with sample_frames set to True, existing frames will not be resampled unless you set force_sample to True.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz
from fiftyone import ViewField as F

dataset = foz.load_zoo_dataset("quickstart-video")

session = fo.launch_app(dataset)

#
# Create a frames view for an entire video dataset
#

frames = dataset.to_frames(sample_frames=True)
print(frames)

session.view = frames

#
# Create a frames view that only contains frames with at least 10
# objects, sampled at a maximum frame rate of 1fps
#

num_objects = F("detections.detections").length()
view = dataset.match_frames(num_objects > 10)

frames = view.to_frames(max_fps=1)
print(frames)

session.view = frames

Parameters:

• sample_frames (False) – whether to assume that the frame images have already been sampled at locations stored in the filepath field of each frame (False), or whether to sample the video frames now according to the specified parameters (True)

• fps (None) – an optional frame rate at which to sample each video’s frames

• max_fps (None) – an optional maximum frame rate at which to sample. Videos with frame rate exceeding this value are downsampled

• size (None) – an optional (width, height) at which to sample frames. A dimension can be -1, in which case the aspect ratio is preserved. Only applicable when sample_frames=True

• min_size (None) – an optional minimum (width, height) for each frame. A dimension can be -1 if no constraint should be applied. The frames are resized (aspect-preserving) if necessary to meet this constraint. Only applicable when sample_frames=True

• max_size (None) – an optional maximum (width, height) for each frame. A dimension can be -1 if no constraint should be applied. The frames are resized (aspect-preserving) if necessary to meet this constraint. Only applicable when sample_frames=True

• sparse (False) – whether to only sample frame images for frame numbers for which fiftyone.core.frame.Frame instances exist in the input collection. This parameter has no effect when sample_frames==False since frames must always exist in order to have filepath information use

• output_dir (None) – an optional output directory in which to write the sampled frames. By default, the frames are written in folders with the same basename of each video

• rel_dir (None) – a relative directory to remove from the filepath of each video, if possible. The path is converted to an absolute path (if necessary) via fiftyone.core.storage.normalize_path(). This argument can be used in conjunction with output_dir to cause the sampled frames to be written in a nested directory structure within output_dir matching the shape of the input video’s folder structure

• frames_patt (None) – a pattern specifying the filename/format to use to write or check or existing sampled frames, e.g., "%%06d.jpg". The default value is fiftyone.config.default_sequence_idx + fiftyone.config.default_image_ext

• force_sample (False) – whether to resample videos whose sampled frames already exist. Only applicable when sample_frames=True

• skip_failures (True) – whether to gracefully continue without raising an error if a video cannot be sampled

• verbose (False) – whether to log information about the frames that will be sampled, if any

• include_indexes (False) – whether to recreate any custom frame indexes on the frames view (True) or a list of specific indexes or index prefixes to recreate. By default, no custom indexes are recreated

Returns:

a fiftyone.core.video.FramesView

classmethod list_aggregations()

Returns a list of all available methods on this collection that apply fiftyone.core.aggregations.Aggregation operations to this collection.

Returns:

a list of SampleCollection method names

bounds(field_or_expr, expr=None, safe=False)

Computes the bounds of a numeric field of the collection.

None-valued fields are ignored.

This aggregation is typically applied to numeric or date field types (or lists of such types):

• fiftyone.core.fields.IntField

• fiftyone.core.fields.FloatField

• fiftyone.core.fields.DateField

• fiftyone.core.fields.DateTimeField

Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            numeric_field=1.0,
            numeric_list_field=[1, 2, 3],
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            numeric_field=4.0,
            numeric_list_field=[1, 2],
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            numeric_field=None,
            numeric_list_field=None,
        ),
    ]
)

#
# Compute the bounds of a numeric field
#

bounds = dataset.bounds("numeric_field")
print(bounds)  # (min, max)

#
# Compute the bounds of a numeric list field
#

bounds = dataset.bounds("numeric_list_field")
print(bounds)  # (min, max)

#
# Compute the bounds of a transformation of a numeric field
#

bounds = dataset.bounds(2 * (F("numeric_field") + 1))
print(bounds)  # (min, max)

Parameters:

• field_or_expr –

  a field name, embedded.field.name, fiftyone.core.expressions.ViewExpression, or MongoDB expression defining the field or expression to aggregate. This can also be a list or tuple of such arguments, in which case a tuple of corresponding aggregation results (each receiving the same additional keyword arguments, if any) will be returned

• expr (None) –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression to apply to field_or_expr (which must be a field) before aggregating

• safe (False) – whether to ignore nan/inf values when dealing with floating point values

Returns:

the (min, max) bounds

count(field_or_expr=None, expr=None, safe=False)

Counts the number of field values in the collection.

None-valued fields are ignored.

If no field is provided, the samples themselves are counted.

Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            predictions=fo.Detections(
                detections=[
                    fo.Detection(label="cat"),
                    fo.Detection(label="dog"),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            predictions=fo.Detections(
                detections=[
                    fo.Detection(label="cat"),
                    fo.Detection(label="rabbit"),
                    fo.Detection(label="squirrel"),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            predictions=None,
        ),
    ]
)

#
# Count the number of samples in the dataset
#

count = dataset.count()
print(count)  # the count

#
# Count the number of samples with `predictions`
#

count = dataset.count("predictions")
print(count)  # the count

#
# Count the number of objects in the `predictions` field
#

count = dataset.count("predictions.detections")
print(count)  # the count

#
# Count the number of objects in samples with > 2 predictions
#

count = dataset.count(
    (F("predictions.detections").length() > 2).if_else(
        F("predictions.detections"), None
    )
)
print(count)  # the count

Parameters:

• field_or_expr (None) –

  a field name, embedded.field.name, fiftyone.core.expressions.ViewExpression, or MongoDB expression defining the field or expression to aggregate. If neither field_or_expr or expr is provided, the samples themselves are counted. This can also be a list or tuple of such arguments, in which case a tuple of corresponding aggregation results (each receiving the same additional keyword arguments, if any) will be returned

• expr (None) –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression to apply to field_or_expr (which must be a field) before aggregating

• safe (False) – whether to ignore nan/inf values when dealing with floating point values

Returns:

the count

count_values(field_or_expr, expr=None, safe=False)

Counts the occurrences of field values in the collection.

This aggregation is typically applied to countable field types (or lists of such types):

• fiftyone.core.fields.BooleanField

• fiftyone.core.fields.IntField

• fiftyone.core.fields.StringField

Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            tags=["sunny"],
            predictions=fo.Detections(
                detections=[
                    fo.Detection(label="cat"),
                    fo.Detection(label="dog"),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            tags=["cloudy"],
            predictions=fo.Detections(
                detections=[
                    fo.Detection(label="cat"),
                    fo.Detection(label="rabbit"),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            predictions=None,
        ),
    ]
)

#
# Compute the tag counts in the dataset
#

counts = dataset.count_values("tags")
print(counts)  # dict mapping values to counts

#
# Compute the predicted label counts in the dataset
#

counts = dataset.count_values("predictions.detections.label")
print(counts)  # dict mapping values to counts

#
# Compute the predicted label counts after some normalization
#

counts = dataset.count_values(
    F("predictions.detections.label").map_values(
        {"cat": "pet", "dog": "pet"}
    ).upper()
)
print(counts)  # dict mapping values to counts

Parameters:

• field_or_expr –

  a field name, embedded.field.name, fiftyone.core.expressions.ViewExpression, or MongoDB expression defining the field or expression to aggregate. This can also be a list or tuple of such arguments, in which case a tuple of corresponding aggregation results (each receiving the same additional keyword arguments, if any) will be returned

• expr (None) –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression to apply to field_or_expr (which must be a field) before aggregating

• safe (False) – whether to treat nan/inf values as None when dealing with floating point values

Returns:

a dict mapping values to counts

distinct(field_or_expr, expr=None, safe=False)

Computes the distinct values of a field in the collection.

None-valued fields are ignored.

This aggregation is typically applied to countable field types (or lists of such types):

• fiftyone.core.fields.BooleanField

• fiftyone.core.fields.IntField

• fiftyone.core.fields.StringField

Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            tags=["sunny"],
            predictions=fo.Detections(
                detections=[
                    fo.Detection(label="cat"),
                    fo.Detection(label="dog"),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            tags=["sunny", "cloudy"],
            predictions=fo.Detections(
                detections=[
                    fo.Detection(label="cat"),
                    fo.Detection(label="rabbit"),
                ]
            ),
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            predictions=None,
        ),
    ]
)

#
# Get the distinct tags in a dataset
#

values = dataset.distinct("tags")
print(values)  # list of distinct values

#
# Get the distinct predicted labels in a dataset
#

values = dataset.distinct("predictions.detections.label")
print(values)  # list of distinct values

#
# Get the distinct predicted labels after some normalization
#

values = dataset.distinct(
    F("predictions.detections.label").map_values(
        {"cat": "pet", "dog": "pet"}
    ).upper()
)
print(values)  # list of distinct values

Parameters:

• field_or_expr –

  a field name, embedded.field.name, fiftyone.core.expressions.ViewExpression, or MongoDB expression defining the field or expression to aggregate. This can also be a list or tuple of such arguments, in which case a tuple of corresponding aggregation results (each receiving the same additional keyword arguments, if any) will be returned

• expr (None) –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression to apply to field_or_expr (which must be a field) before aggregating

• safe (False) – whether to ignore nan/inf values when dealing with floating point values

Returns:

a sorted list of distinct values

histogram_values(field_or_expr, expr=None, bins=None, range=None, auto=False)

Computes a histogram of the field values in the collection.

This aggregation is typically applied to numeric field types (or lists of such types):

• fiftyone.core.fields.IntField

• fiftyone.core.fields.FloatField

Examples:

import numpy as np
import matplotlib.pyplot as plt

import fiftyone as fo
from fiftyone import ViewField as F

samples = []
for idx in range(100):
    samples.append(
        fo.Sample(
            filepath="/path/to/image%d.png" % idx,
            numeric_field=np.random.randn(),
            numeric_list_field=list(np.random.randn(10)),
        )
    )

dataset = fo.Dataset()
dataset.add_samples(samples)

def plot_hist(counts, edges):
    counts = np.asarray(counts)
    edges = np.asarray(edges)
    left_edges = edges[:-1]
    widths = edges[1:] - edges[:-1]
    plt.bar(left_edges, counts, width=widths, align="edge")

#
# Compute a histogram of a numeric field
#

counts, edges, other = dataset.histogram_values(
    "numeric_field", bins=50, range=(-4, 4)
)

plot_hist(counts, edges)
plt.show(block=False)

#
# Compute the histogram of a numeric list field
#

counts, edges, other = dataset.histogram_values(
    "numeric_list_field", bins=50
)

plot_hist(counts, edges)
plt.show(block=False)

#
# Compute the histogram of a transformation of a numeric field
#

counts, edges, other = dataset.histogram_values(
    2 * (F("numeric_field") + 1), bins=50
)

plot_hist(counts, edges)
plt.show(block=False)

Parameters:

• field_or_expr –

  a field name, embedded.field.name, fiftyone.core.expressions.ViewExpression, or MongoDB expression defining the field or expression to aggregate. This can also be a list or tuple of such arguments, in which case a tuple of corresponding aggregation results (each receiving the same additional keyword arguments, if any) will be returned

• expr (None) –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression to apply to field_or_expr (which must be a field) before aggregating

• bins (None) – can be either an integer number of bins to generate or a monotonically increasing sequence specifying the bin edges to use. By default, 10 bins are created. If bins is an integer and no range is specified, bin edges are automatically distributed in an attempt to evenly distribute the counts in each bin

• range (None) – a (lower, upper) tuple specifying a range in which to generate equal-width bins. Only applicable when bins is an integer

• auto (False) – whether to automatically choose bin edges in an attempt to evenly distribute the counts in each bin. If this option is chosen, bins will only be used if it is an integer, and the range parameter is ignored

Returns:

a tuple of

• counts: a list of counts in each bin

• edges: an increasing list of bin edges of length len(counts) + 1. Note that each bin is treated as having an inclusive lower boundary and exclusive upper boundary, [lower, upper), including the rightmost bin

• other: the number of items outside the bins

min(field_or_expr, expr=None, safe=False)

Computes the minimum of a numeric field of the collection.

None-valued fields are ignored.

This aggregation is typically applied to numeric or date field types (or lists of such types):

• fiftyone.core.fields.IntField

• fiftyone.core.fields.FloatField

• fiftyone.core.fields.DateField

• fiftyone.core.fields.DateTimeField

Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            numeric_field=1.0,
            numeric_list_field=[1, 2, 3],
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            numeric_field=4.0,
            numeric_list_field=[1, 2],
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            numeric_field=None,
            numeric_list_field=None,
        ),
    ]
)

#
# Compute the minimum of a numeric field
#

min = dataset.min("numeric_field")
print(min)  # the min

#
# Compute the minimum of a numeric list field
#

min = dataset.min("numeric_list_field")
print(min)  # the min

#
# Compute the minimum of a transformation of a numeric field
#

min = dataset.min(2 * (F("numeric_field") + 1))
print(min)  # the min

Parameters:

• field_or_expr –

  a field name, embedded.field.name, fiftyone.core.expressions.ViewExpression, or MongoDB expression defining the field or expression to aggregate. This can also be a list or tuple of such arguments, in which case a tuple of corresponding aggregation results (each receiving the same additional keyword arguments, if any) will be returned

• expr (None) –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression to apply to field_or_expr (which must be a field) before aggregating

• safe (False) – whether to ignore nan/inf values when dealing with floating point values

Returns:

the minimum value

max(field_or_expr, expr=None, safe=False)

Computes the maximum of a numeric field of the collection.

None-valued fields are ignored.

This aggregation is typically applied to numeric or date field types (or lists of such types):

• fiftyone.core.fields.IntField

• fiftyone.core.fields.FloatField

• fiftyone.core.fields.DateField

• fiftyone.core.fields.DateTimeField

Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            numeric_field=1.0,
            numeric_list_field=[1, 2, 3],
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            numeric_field=4.0,
            numeric_list_field=[1, 2],
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            numeric_field=None,
            numeric_list_field=None,
        ),
    ]
)

#
# Compute the maximum of a numeric field
#

max = dataset.max("numeric_field")
print(max)  # the max

#
# Compute the maximum of a numeric list field
#

max = dataset.max("numeric_list_field")
print(max)  # the max

#
# Compute the maximum of a transformation of a numeric field
#

max = dataset.max(2 * (F("numeric_field") + 1))
print(max)  # the max

Parameters:

• field_or_expr –

  a field name, embedded.field.name, fiftyone.core.expressions.ViewExpression, or MongoDB expression defining the field or expression to aggregate. This can also be a list or tuple of such arguments, in which case a tuple of corresponding aggregation results (each receiving the same additional keyword arguments, if any) will be returned

• expr (None) –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression to apply to field_or_expr (which must be a field) before aggregating

• safe (False) – whether to ignore nan/inf values when dealing with floating point values

Returns:

the maximum value

mean(field_or_expr, expr=None, safe=False)

Computes the arithmetic mean of the field values of the collection.

None-valued fields are ignored.

This aggregation is typically applied to numeric field types (or lists of such types):

• fiftyone.core.fields.IntField

• fiftyone.core.fields.FloatField

Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            numeric_field=1.0,
            numeric_list_field=[1, 2, 3],
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            numeric_field=4.0,
            numeric_list_field=[1, 2],
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            numeric_field=None,
            numeric_list_field=None,
        ),
    ]
)

#
# Compute the mean of a numeric field
#

mean = dataset.mean("numeric_field")
print(mean)  # the mean

#
# Compute the mean of a numeric list field
#

mean = dataset.mean("numeric_list_field")
print(mean)  # the mean

#
# Compute the mean of a transformation of a numeric field
#

mean = dataset.mean(2 * (F("numeric_field") + 1))
print(mean)  # the mean

Parameters:

• field_or_expr –

  a field name, embedded.field.name, fiftyone.core.expressions.ViewExpression, or MongoDB expression defining the field or expression to aggregate. This can also be a list or tuple of such arguments, in which case a tuple of corresponding aggregation results (each receiving the same additional keyword arguments, if any) will be returned

• expr (None) –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression to apply to field_or_expr (which must be a field) before aggregating

• safe (False) – whether to ignore nan/inf values when dealing with floating point values

Returns:

the mean

quantiles(field_or_expr, quantiles, expr=None, safe=False)

Computes the quantile(s) of the field values of a collection.

None-valued fields are ignored.

This aggregation is typically applied to numeric field types (or lists of such types):

• fiftyone.core.fields.IntField

• fiftyone.core.fields.FloatField

Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            numeric_field=1.0,
            numeric_list_field=[1, 2, 3],
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            numeric_field=4.0,
            numeric_list_field=[1, 2],
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            numeric_field=None,
            numeric_list_field=None,
        ),
    ]
)

#
# Compute the quantiles of a numeric field
#

quantiles = dataset.quantiles("numeric_field", [0.1, 0.5, 0.9])
print(quantiles)  # the quantiles

#
# Compute the quantiles of a numeric list field
#

quantiles = dataset.quantiles("numeric_list_field", [0.1, 0.5, 0.9])
print(quantiles)  # the quantiles

#
# Compute the mean of a transformation of a numeric field
#

quantiles = dataset.quantiles(2 * (F("numeric_field") + 1), [0.1, 0.5, 0.9])
print(quantiles)  # the quantiles

Parameters:

• field_or_expr –

  a field name, embedded.field.name, fiftyone.core.expressions.ViewExpression, or MongoDB expression defining the field or expression to aggregate

• quantiles – the quantile or iterable of quantiles to compute. Each quantile must be a numeric value in [0, 1]

• expr (None) –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression to apply to field_or_expr (which must be a field) before aggregating

• safe (False) – whether to ignore nan/inf values when dealing with floating point values

Returns:

the quantile or list of quantiles

schema(field_or_expr, expr=None, dynamic_only=False, _doc_type=None, _include_private=False)

Extracts the names and types of the attributes of a specified embedded document field across all samples in the collection.

Schema aggregations are useful for detecting the presence and types of dynamic attributes of fiftyone.core.labels.Label fields across a collection.

Examples:

import fiftyone as fo

dataset = fo.Dataset()

sample1 = fo.Sample(
    filepath="image1.png",
    ground_truth=fo.Detections(
        detections=[
            fo.Detection(
                label="cat",
                bounding_box=[0.1, 0.1, 0.4, 0.4],
                foo="bar",
                hello=True,
            ),
            fo.Detection(
                label="dog",
                bounding_box=[0.5, 0.5, 0.4, 0.4],
                hello=None,
            )
        ]
    )
)

sample2 = fo.Sample(
    filepath="image2.png",
    ground_truth=fo.Detections(
        detections=[
            fo.Detection(
                label="rabbit",
                bounding_box=[0.1, 0.1, 0.4, 0.4],
                foo=None,
            ),
            fo.Detection(
                label="squirrel",
                bounding_box=[0.5, 0.5, 0.4, 0.4],
                hello="there",
            ),
        ]
    )
)

dataset.add_samples([sample1, sample2])

#
# Get schema of all dynamic attributes on the detections in a
# `Detections` field
#

print(dataset.schema("ground_truth.detections", dynamic_only=True))
# {'foo': StringField, 'hello': [BooleanField, StringField]}

Parameters:

• field_or_expr –

  a field name, embedded.field.name, fiftyone.core.expressions.ViewExpression, or MongoDB expression defining the field or expression to aggregate

• expr (None) –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression to apply to field_or_expr (which must be a field) before aggregating

• dynamic_only (False) – whether to only include dynamically added attributes

Returns:

a dict mapping field names to fiftyone.core.fields.Field instances. If a field’s values takes multiple non-None types, the list of observed types will be returned

list_schema(field_or_expr, expr=None)

Extracts the value type(s) in a specified list field across all samples in the collection.

Examples:

from datetime import datetime
import fiftyone as fo

dataset = fo.Dataset()

sample1 = fo.Sample(
    filepath="image1.png",
    ground_truth=fo.Classification(
        label="cat",
        info=[
            fo.DynamicEmbeddedDocument(
                task="initial_annotation",
                author="Alice",
                timestamp=datetime(1970, 1, 1),
                notes=["foo", "bar"],
            ),
            fo.DynamicEmbeddedDocument(
                task="editing_pass",
                author="Bob",
                timestamp=datetime.utcnow(),
            ),
        ],
    ),
)

sample2 = fo.Sample(
    filepath="image2.png",
    ground_truth=fo.Classification(
        label="dog",
        info=[
            fo.DynamicEmbeddedDocument(
                task="initial_annotation",
                author="Bob",
                timestamp=datetime(2018, 10, 18),
                notes=["spam", "eggs"],
            ),
        ],
    ),
)

dataset.add_samples([sample1, sample2])

# Determine that `ground_truth.info` contains embedded documents
print(dataset.list_schema("ground_truth.info"))
# fo.EmbeddedDocumentField

# Determine the fields of the embedded documents in the list
print(dataset.schema("ground_truth.info[]"))
# {'task': StringField, ..., 'notes': ListField}

# Determine the type of the values in the nested `notes` list field
# Since `ground_truth.info` is not yet declared on the dataset's
# schema, we must manually include `[]` to unwind the info lists
print(dataset.list_schema("ground_truth.info[].notes"))
# fo.StringField

# Declare the `ground_truth.info` field
dataset.add_sample_field(
    "ground_truth.info",
    fo.ListField,
    subfield=fo.EmbeddedDocumentField,
    embedded_doc_type=fo.DynamicEmbeddedDocument,
)

# Now we can inspect the nested `notes` field without unwinding
print(dataset.list_schema("ground_truth.info.notes"))
# fo.StringField

Parameters:

• field_or_expr –

  a field name, embedded.field.name, fiftyone.core.expressions.ViewExpression, or MongoDB expression defining the field or expression to aggregate

• expr (None) –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression to apply to field_or_expr (which must be a field) before aggregating

Returns:

a fiftyone.core.fields.Field or list of fiftyone.core.fields.Field instances describing the value type(s) in the list

std(field_or_expr, expr=None, safe=False, sample=False)

Computes the standard deviation of the field values of the collection.

None-valued fields are ignored.

This aggregation is typically applied to numeric field types (or lists of such types):

• fiftyone.core.fields.IntField

• fiftyone.core.fields.FloatField

Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            numeric_field=1.0,
            numeric_list_field=[1, 2, 3],
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            numeric_field=4.0,
            numeric_list_field=[1, 2],
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            numeric_field=None,
            numeric_list_field=None,
        ),
    ]
)

#
# Compute the standard deviation of a numeric field
#

std = dataset.std("numeric_field")
print(std)  # the standard deviation

#
# Compute the standard deviation of a numeric list field
#

std = dataset.std("numeric_list_field")
print(std)  # the standard deviation

#
# Compute the standard deviation of a transformation of a numeric field
#

std = dataset.std(2 * (F("numeric_field") + 1))
print(std)  # the standard deviation

Parameters:

• field_or_expr –

  a field name, embedded.field.name, fiftyone.core.expressions.ViewExpression, or MongoDB expression defining the field or expression to aggregate. This can also be a list or tuple of such arguments, in which case a tuple of corresponding aggregation results (each receiving the same additional keyword arguments, if any) will be returned

• expr (None) –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression to apply to field_or_expr (which must be a field) before aggregating

• safe (False) – whether to ignore nan/inf values when dealing with floating point values

• sample (False) – whether to compute the sample standard deviation rather than the population standard deviation

Returns:

the standard deviation

sum(field_or_expr, expr=None, safe=False)

Computes the sum of the field values of the collection.

None-valued fields are ignored.

This aggregation is typically applied to numeric field types (or lists of such types):

• fiftyone.core.fields.IntField

• fiftyone.core.fields.FloatField

Examples:

import fiftyone as fo
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            numeric_field=1.0,
            numeric_list_field=[1, 2, 3],
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            numeric_field=4.0,
            numeric_list_field=[1, 2],
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            numeric_field=None,
            numeric_list_field=None,
        ),
    ]
)

#
# Compute the sum of a numeric field
#

total = dataset.sum("numeric_field")
print(total)  # the sum

#
# Compute the sum of a numeric list field
#

total = dataset.sum("numeric_list_field")
print(total)  # the sum

#
# Compute the sum of a transformation of a numeric field
#

total = dataset.sum(2 * (F("numeric_field") + 1))
print(total)  # the sum

Parameters:

• field_or_expr –

  a field name, embedded.field.name, fiftyone.core.expressions.ViewExpression, or MongoDB expression defining the field or expression to aggregate. This can also be a list or tuple of such arguments, in which case a tuple of corresponding aggregation results (each receiving the same additional keyword arguments, if any) will be returned

• expr (None) –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression to apply to field_or_expr (which must be a field) before aggregating

• safe (False) – whether to ignore nan/inf values when dealing with floating point values

Returns:

the sum

values(field_or_expr, expr=None, missing_value=None, unwind=False, _allow_missing=False, _big_result=True, _raw=False, _field=None, _enforce_natural_order=True)

Extracts the values of a field from all samples in the collection.

Values aggregations are useful for efficiently extracting a slice of field or embedded field values across all samples in a collection. See the examples below for more details.

The dual function of values() is set_values(), which can be used to efficiently set a field or embedded field of all samples in a collection by providing lists of values of same structure returned by this aggregation.

Note

Unlike other aggregations, values() does not automatically unwind list fields, which ensures that the returned values match the potentially-nested structure of the documents.

You can opt-in to unwinding specific list fields using the [] syntax, or you can pass the optional unwind=True parameter to unwind all supported list fields. See Aggregating list fields for more information.

Examples:

import fiftyone as fo
import fiftyone.zoo as foz
from fiftyone import ViewField as F

dataset = fo.Dataset()
dataset.add_samples(
    [
        fo.Sample(
            filepath="/path/to/image1.png",
            numeric_field=1.0,
            numeric_list_field=[1, 2, 3],
        ),
        fo.Sample(
            filepath="/path/to/image2.png",
            numeric_field=4.0,
            numeric_list_field=[1, 2],
        ),
        fo.Sample(
            filepath="/path/to/image3.png",
            numeric_field=None,
            numeric_list_field=None,
        ),
    ]
)

#
# Get all values of a field
#

values = dataset.values("numeric_field")
print(values)  # [1.0, 4.0, None]

#
# Get all values of a list field
#

values = dataset.values("numeric_list_field")
print(values)  # [[1, 2, 3], [1, 2], None]

#
# Get all values of transformed field
#

values = dataset.values(2 * (F("numeric_field") + 1))
print(values)  # [4.0, 10.0, None]

#
# Get values from a label list field
#

dataset = foz.load_zoo_dataset("quickstart")

# list of `Detections`
detections = dataset.values("ground_truth")

# list of lists of `Detection` instances
detections = dataset.values("ground_truth.detections")

# list of lists of detection labels
labels = dataset.values("ground_truth.detections.label")

Parameters:

• field_or_expr –

  a field name, embedded.field.name, fiftyone.core.expressions.ViewExpression, or MongoDB expression defining the field or expression to aggregate. This can also be a list or tuple of such arguments, in which case a tuple of corresponding aggregation results (each receiving the same additional keyword arguments, if any) will be returned

• expr (None) –

  a fiftyone.core.expressions.ViewExpression or MongoDB expression to apply to field_or_expr (which must be a field) before aggregating

• missing_value (None) – a value to insert for missing or None-valued fields

• unwind (False) – whether to automatically unwind all recognized list fields (True) or unwind all list fields except the top-level sample field (-1)

Returns:

the list of values

draw_labels(output_dir, rel_dir=None, label_fields=None, overwrite=False, config=None, progress=None, **kwargs)

Renders annotated versions of the media in the collection with the specified label data overlaid to the given directory.

The filenames of the sample media are maintained, unless a name conflict would occur in output_dir, in which case an index of the form "-%d" % count is appended to the base filename.

Images are written in format fo.config.default_image_ext, and videos are written in format fo.config.default_video_ext.

Parameters:

• output_dir – the directory to write the annotated media

• rel_dir (None) – an optional relative directory to strip from each input filepath to generate a unique identifier that is joined with output_dir to generate an output path for each annotated media. This argument allows for populating nested subdirectories in output_dir that match the shape of the input paths. The path is converted to an absolute path (if necessary) via fiftyone.core.storage.normalize_path()

• label_fields (None) – a label field or list of label fields to render. By default, all fiftyone.core.labels.Label fields are drawn

• overwrite (False) – whether to delete output_dir if it exists before rendering

• config (None) – an optional fiftyone.utils.annotations.DrawConfig configuring how to draw the labels

• progress (None) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead

• **kwargs – optional keyword arguments specifying parameters of the default fiftyone.utils.annotations.DrawConfig to override

Returns:

the list of paths to the rendered media

export(export_dir=None, dataset_type=None, data_path=None, labels_path=None, export_media=None, rel_dir=None, dataset_exporter=None, label_field=None, frame_labels_field=None, overwrite=False, progress=None, **kwargs)

Exports the samples in the collection to disk.

You can perform exports with this method via the following basic patterns:

1. Provide export_dir and dataset_type to export the content to a directory in the default layout for the specified format, as documented in this page

2. Provide dataset_type along with data_path, labels_path, and/or export_media to directly specify where to export the source media and/or labels (if applicable) in your desired format. This syntax provides the flexibility to, for example, perform workflows like labels-only exports

3. Provide a dataset_exporter to which to feed samples to perform a fully-customized export

In all workflows, the remaining parameters of this method can be provided to further configure the export.

See this page for more information about the available export formats and examples of using this method.

See this guide for more details about exporting datasets in custom formats by defining your own fiftyone.utils.data.exporters.DatasetExporter.

This method will automatically coerce the data to match the requested export in the following cases:

• When exporting in either an unlabeled image or image classification format, if a spatial label field is provided (fiftyone.core.labels.Detection, fiftyone.core.labels.Detections, fiftyone.core.labels.Polyline, or fiftyone.core.labels.Polylines), then the image patches of the provided samples will be exported

• When exporting in labeled image dataset formats that expect list-type labels (fiftyone.core.labels.Classifications, fiftyone.core.labels.Detections, fiftyone.core.labels.Keypoints, or fiftyone.core.labels.Polylines), if a label field contains labels in non-list format (e.g., fiftyone.core.labels.Classification), the labels will be automatically upgraded to single-label lists

• When exporting in labeled image dataset formats that expect fiftyone.core.labels.Detections labels, if a fiftyone.core.labels.Classification field is provided, the labels will be automatically upgraded to detections that span the entire images

Parameters:

• export_dir (None) –

  the directory to which to export the samples in format dataset_type. This parameter may be omitted if you have provided appropriate values for the data_path and/or labels_path parameters. Alternatively, this can also be an archive path with one of the following extensions:

  .zip, .tar, .tar.gz, .tgz, .tar.bz, .tbz
  

  If an archive path is specified, the export is performed in a directory of same name (minus extension) and then automatically archived and the directory then deleted

• dataset_type (None) – the fiftyone.types.Dataset type to write. If not specified, the default type for label_field is used

• data_path (None) –

  an optional parameter that enables explicit control over the location of the exported media for certain export formats. Can be any of the following:

  • a folder name like "data" or "data/" specifying a subfolder of export_dir in which to export the media

  • an absolute directory path in which to export the media. In this case, the export_dir has no effect on the location of the data

  • a filename like "data.json" specifying the filename of a JSON manifest file in export_dir generated when export_media is "manifest"

  • an absolute filepath specifying the location to write the JSON manifest file when export_media is "manifest". In this case, export_dir has no effect on the location of the data

  If None, a default value of this parameter will be chosen based on the value of the export_media parameter. Note that this parameter is not applicable to certain export formats such as binary types like TF records

• labels_path (None) –

  an optional parameter that enables explicit control over the location of the exported labels. Only applicable when exporting in certain labeled dataset formats. Can be any of the following:

  • a type-specific folder name like "labels" or "labels/" or a filename like "labels.json" or "labels.xml" specifying the location in export_dir in which to export the labels

  • an absolute directory or filepath in which to export the labels. In this case, the export_dir has no effect on the location of the labels

  For labeled datasets, the default value of this parameter will be chosen based on the export format so that the labels will be exported into export_dir

• export_media (None) –

  controls how to export the raw media. The supported values are:

  • True: copy all media files into the output directory

  • False: don’t export media. This option is only useful when exporting labeled datasets whose label format stores sufficient information to locate the associated media

  • "move": move all media files into the output directory

  • "symlink": create symlinks to the media files in the output directory

  • "manifest": create a data.json in the output directory that maps UUIDs used in the labels files to the filepaths of the source media, rather than exporting the actual media

  If None, an appropriate default value of this parameter will be chosen based on the value of the data_path parameter. Note that some dataset formats may not support certain values for this parameter (e.g., when exporting in binary formats such as TF records, “symlink” is not an option)

• rel_dir (None) – an optional relative directory to strip from each input filepath to generate a unique identifier for each media. When exporting media, this identifier is joined with data_path to generate an output path for each exported media. This argument allows for populating nested subdirectories that match the shape of the input paths. The path is converted to an absolute path (if necessary) via fiftyone.core.storage.normalize_path()

• dataset_exporter (None) – a fiftyone.utils.data.exporters.DatasetExporter to use to export the samples. When provided, parameters such as export_dir, dataset_type, data_path, and labels_path have no effect

• label_field (None) –

  controls the label field(s) to export. Only applicable to labeled datasets. Can be any of the following:

  • the name of a label field to export

  • a glob pattern of label field(s) to export

  • a list or tuple of label field(s) to export

  • a dictionary mapping label field names to keys to use when constructing the label dictionaries to pass to the exporter

  Note that multiple fields can only be specified when the exporter used can handle dictionaries of labels. By default, the first field of compatible type for the exporter is used. When exporting labeled video datasets, this argument may contain frame fields prefixed by "frames."

• frame_labels_field (None) –

  controls the frame label field(s) to export. The "frames." prefix is optional. Only applicable to labeled video datasets. Can be any of the following:

  • the name of a frame label field to export

  • a glob pattern of frame label field(s) to export

  • a list or tuple of frame label field(s) to export

  • a dictionary mapping frame label field names to keys to use when constructing the frame label dictionaries to pass to the exporter

  Note that multiple fields can only be specified when the exporter used can handle dictionaries of frame labels. By default, the first field of compatible type for the exporter is used

• overwrite (False) – whether to delete existing directories before performing the export (True) or to merge the export with existing files and directories (False)

• progress (None) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead

• **kwargs – optional keyword arguments to pass to the dataset exporter’s constructor. If you are exporting image patches, this can also contain keyword arguments for fiftyone.utils.patches.ImagePatchesExtractor

to_torch(get_item, vectorize=False, skip_failures=False, local_process_group=None)

Constructs a torch.utils.data.Dataset that loads data from this collection via the provided fiftyone.utils.torch.GetItem instance.

Parameters:

• get_item – a fiftyone.utils.torch.GetItem

• vectorize (False) – whether to load and cache the required fields from the sample collection upfront (True) or lazily load the values from each sample when items are retrieved (False). Vectorizing gives faster data loading times, but you must have enough memory to store the required field values for the entire sample collection. When vectorize=True, all field values must be serializable; ie pickle.dumps(field_value) must not raise an error

• skip_failures (False) – whether to skip failures that occur when calling get_item. If True, the exception will be returned rather than the intended field values

• local_process_group (None) – the local process group. Only used during distributed training

Returns:

a torch.utils.data.Dataset

annotate(anno_key, label_schema=None, label_field=None, label_type=None, classes=None, attributes=True, mask_targets=None, allow_additions=True, allow_deletions=True, allow_label_edits=True, allow_index_edits=True, allow_spatial_edits=True, media_field='filepath', backend=None, launch_editor=False, **kwargs)

Exports the samples and optional label field(s) in this collection to the given annotation backend.

The backend parameter controls which annotation backend to use. Depending on the backend you use, you may want/need to provide extra keyword arguments to this function for the constructor of the backend’s fiftyone.utils.annotations.AnnotationBackendConfig class.

The natively provided backends and their associated config classes are:

• "cvat": fiftyone.utils.cvat.CVATBackendConfig

• "labelstudio": fiftyone.utils.labelstudio.LabelStudioBackendConfig

• "labelbox": fiftyone.utils.labelbox.LabelboxBackendConfig

See this page for more information about using this method, including how to define label schemas and how to configure login credentials for your annotation provider.

Parameters:

• anno_key – a string key to use to refer to this annotation run

• label_schema (None) – a dictionary defining the label schema to use. If this argument is provided, it takes precedence over the other schema-related arguments

• label_field (None) – a string indicating a new or existing label field to annotate

• label_type (None) –

  a string indicating the type of labels to annotate. The possible values are:

  • "classification": a single classification stored in fiftyone.core.labels.Classification fields

  • "classifications": multilabel classifications stored in fiftyone.core.labels.Classifications fields

  • "detections": object detections stored in fiftyone.core.labels.Detections fields

  • "instances": instance segmentations stored in fiftyone.core.labels.Detections fields with their mask attributes populated

  • "polylines": polylines stored in fiftyone.core.labels.Polylines fields with their filled attributes set to False

  • "polygons": polygons stored in fiftyone.core.labels.Polylines fields with their filled attributes set to True

  • "keypoints": keypoints stored in fiftyone.core.labels.Keypoints fields

  • "segmentation": semantic segmentations stored in fiftyone.core.labels.Segmentation fields

  • "scalar": scalar labels stored in fiftyone.core.fields.IntField, fiftyone.core.fields.FloatField, fiftyone.core.fields.StringField, or fiftyone.core.fields.BooleanField fields

  All new label fields must have their type specified via this argument or in label_schema. Note that annotation backends may not support all label types

• classes (None) – a list of strings indicating the class options for label_field or all fields in label_schema without classes specified. All new label fields must have a class list provided via one of the supported methods. For existing label fields, if classes are not provided by this argument nor label_schema, they are retrieved from get_classes() if possible, or else the observed labels on your dataset are used

• attributes (True) –

  specifies the label attributes of each label field to include (other than their label, which is always included) in the annotation export. Can be any of the following:

  • True: export all label attributes

  • False: don’t export any custom label attributes

  • a list of label attributes to export

  • a dict mapping attribute names to dicts specifying the type, values, and default for each attribute

  If a label_schema is also provided, this parameter determines which attributes are included for all fields that do not explicitly define their per-field attributes (in addition to any per-class attributes)

• mask_targets (None) – a dict mapping pixel values to semantic label strings. Only applicable when annotating semantic segmentations

• allow_additions (True) – whether to allow new labels to be added. Only applicable when editing existing label fields

• allow_deletions (True) – whether to allow labels to be deleted. Only applicable when editing existing label fields

• allow_label_edits (True) – whether to allow the label attribute of existing labels to be modified. Only applicable when editing existing fields with label attributes

• allow_index_edits (True) – whether to allow the index attribute of existing video tracks to be modified. Only applicable when editing existing frame fields with index attributes

• allow_spatial_edits (True) – whether to allow edits to the spatial properties (bounding boxes, vertices, keypoints, masks, etc) of labels. Only applicable when editing existing spatial label fields

• media_field ("filepath") – the field containing the paths to the media files to upload

• backend (None) – the annotation backend to use. The supported values are fiftyone.annotation_config.backends.keys() and the default is fiftyone.annotation_config.default_backend

• launch_editor (False) – whether to launch the annotation backend’s editor after uploading the samples

• **kwargs – keyword arguments for the fiftyone.utils.annotations.AnnotationBackendConfig

Returns:

an fiftyone.utils.annotations.AnnnotationResults

property has_annotation_runs

Whether this collection has any annotation runs.

has_annotation_run(anno_key)

Whether this collection has an annotation run with the given key.

Parameters:

anno_key – an annotation key

Returns:

True/False

list_annotation_runs(type=None, method=None, **kwargs)

Returns a list of annotation keys on this collection.

Parameters:

• type (None) –

  a specific annotation run type to match, which can be:

  • a string fiftyone.core.annotations.AnnotationMethodConfig.type

  • a fiftyone.core.annotations.AnnotationMethod class or its fully-qualified class name string

• method (None) – a specific fiftyone.core.annotations.AnnotationMethodConfig.method string to match

• **kwargs – optional config parameters to match

Returns:

a list of annotation keys

rename_annotation_run(anno_key, new_anno_key)

Replaces the key for the given annotation run with a new key.

Parameters:

• anno_key – an annotation key

• new_anno_key – a new annotation key

get_annotation_info(anno_key)

Returns information about the annotation run with the given key on this collection.

Parameters:

anno_key – an annotation key

Returns:

a fiftyone.core.annotation.AnnotationInfo

load_annotation_results(anno_key, cache=True, **kwargs)

Loads the results for the annotation run with the given key on this collection.

The fiftyone.utils.annotations.AnnotationResults object returned by this method will provide a variety of backend-specific methods allowing you to perform actions such as checking the status and deleting this run from the annotation backend.

Use load_annotations() to load the labels from an annotation run onto your FiftyOne dataset.

Parameters:

• anno_key – an annotation key

• cache (True) – whether to cache the results on the collection

• **kwargs – keyword arguments for run’s fiftyone.core.annotation.AnnotationMethodConfig.load_credentials() method

Returns:

a fiftyone.utils.annotations.AnnotationResults

load_annotation_view(anno_key, select_fields=False)

Loads the fiftyone.core.view.DatasetView on which the specified annotation run was performed on this collection.

Parameters:

• anno_key – an annotation key

• select_fields (False) – whether to exclude fields involved in other annotation runs

Returns:

a fiftyone.core.view.DatasetView

load_annotations(anno_key, dest_field=None, unexpected='prompt', cleanup=False, progress=None, **kwargs)

Downloads the labels from the given annotation run from the annotation backend and merges them into this collection.

See this page for more information about using this method to import annotations that you have scheduled by calling annotate().

Parameters:

• anno_key – an annotation key

• dest_field (None) – an optional name of a new destination field into which to load the annotations, or a dict mapping field names in the run’s label schema to new destination field names

• unexpected ("prompt") –

  how to deal with any unexpected labels that don’t match the run’s label schema when importing. The supported values are:

  • "prompt": present an interactive prompt to direct/discard unexpected labels

  • "ignore": automatically ignore any unexpected labels

  • "keep": automatically keep all unexpected labels in a field whose name matches the the label type

  • "return": return a dict containing all unexpected labels, or None if there aren’t any

• cleanup (False) – whether to delete any informtation regarding this run from the annotation backend after loading the annotations

• progress (None) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead

• **kwargs – keyword arguments for the run’s fiftyone.core.annotation.AnnotationMethodConfig.load_credentials() method

Returns:

None, unless unexpected=="return" and unexpected labels are found, in which case a dict containing the extra labels is returned

delete_annotation_run(anno_key)

Deletes the annotation run with the given key from this collection.

Calling this method only deletes the record of the annotation run from the collection; it will not delete any annotations loaded onto your dataset via load_annotations(), nor will it delete any associated information from the annotation backend.

Use load_annotation_results() to programmatically manage/delete a run from the annotation backend.

Parameters:

anno_key – an annotation key

delete_annotation_runs()

Deletes all annotation runs from this collection.

Calling this method only deletes the records of the annotation runs from this collection; it will not delete any annotations loaded onto your dataset via load_annotations(), nor will it delete any associated information from the annotation backend.

Use load_annotation_results() to programmatically manage/delete runs in the annotation backend.

list_indexes()

Returns the list of index names on this collection.

Single-field indexes are referenced by their field name, while compound indexes are referenced by more complicated strings. See pymongo.collection.Collection.index_information() for details on the compound format.

Returns:

the list of index names

get_index_information(include_stats=False, _keep_index_names=False)

Returns a dictionary of information about the indexes on this collection.

See pymongo.collection.Collection.index_information() for details on the structure of this dictionary.

Parameters:

include_stats (False) – whether to include the size, usage, and build status of each index

Returns:

a dict mapping index names to info dicts

create_index(field_or_spec, unique=False, wait=True, **kwargs)

Creates an index on the given field or with the given specification, if necessary.

Indexes enable efficient sorting, merging, and other such operations.

Frame-level fields can be indexed by prepending "frames." to the field name.

Note

If an index with the same field(s) but different order(s) already exists, no new index will be created.

Use drop_index() to drop an existing index first if you wish to replace an existing index with new properties.

Note

If you are indexing a single field and it already has a unique constraint, it will be retained regardless of the unique value you specify. Conversely, if the given field already has a non-unique index but you requested a unique index, the existing index will be replaced with a unique index.

Use drop_index() to drop an existing index first if you wish to replace an existing index with new properties.

Parameters:

• field_or_spec – the field name, embedded.field.name, or index specification list. See pymongo.collection.Collection.create_index() for supported values

• unique (False) – whether to add a uniqueness constraint to the index

• wait (True) – whether to wait for index creation to finish

• **kwargs – optional keyword arguments for pymongo.collection.Collection.create_index()

Returns:

the name of the index

drop_index(field_or_name)

Drops the index for the given field or name, if necessary.

Parameters:

field_or_name – a field name, embedded.field.name, or compound index name. Use list_indexes() to see the available indexes

reload()

Reloads the collection from the database.

to_dict(rel_dir=None, include_private=False, include_frames=False, frame_labels_dir=None, pretty_print=False, progress=None)

Returns a JSON dictionary representation of the collection.

Parameters:

• rel_dir (None) – a relative directory to remove from the filepath of each sample, if possible. The path is converted to an absolute path (if necessary) via fiftyone.core.storage.normalize_path(). The typical use case for this argument is that your source data lives in a single directory and you wish to serialize relative, rather than absolute, paths to the data within that directory

• include_private (False) – whether to include private fields

• include_frames (False) – whether to include the frame labels for video samples

• frame_labels_dir (None) – a directory in which to write per-sample JSON files containing the frame labels for video samples. If omitted, frame labels will be included directly in the returned JSON dict (which can be quite quite large for video datasets containing many frames). Only applicable to datasets that contain videos when include_frames is True

• pretty_print (False) – whether to render frame labels JSON in human readable format with newlines and indentations. Only applicable to datasets that contain videos when a frame_labels_dir is provided

• progress (None) – whether to render a progress bar (True/False), use the default value fiftyone.config.show_progress_bars (None), or a progress callback function to invoke instead

Returns:

a JSON dict

to_json(rel_dir=None, include_private=False, include_frames=False, frame_labels_dir=None, pretty_print=False)

Returns a JSON string representation of the collection.

The samples will be written as a list in a top-level samples field of the returned dictionary.

Parameters:

• rel_dir (None) – a relative directory to remove from the filepath of each sample, if possible. The path is converted to an absolute path (if necessary) via fiftyone.core.storage.normalize_path(). The typical use case for this argument is that your source data lives in a single directory and you wish to serialize relative, rather than absolute, paths to the data within that directory

• include_private (False) – whether to include private fields

• include_frames (False) – whether to include the frame labels for video samples

• frame_labels_dir (None) – a directory in which to write per-sample JSON files containing the frame labels for video samples. If omitted, frame labels will be included directly in the returned JSON dict (which can be quite quite large for video datasets containing many frames). Only applicable to datasets that contain videos when include_frames is True

• pretty_print (False) – whether to render the JSON in human readable format with newlines and indentations

Returns:

a JSON string

write_json(json_path, rel_dir=None, include_private=False, include_frames=False, frame_labels_dir=None, pretty_print=False)

Writes the colllection to disk in JSON format.

Parameters:

• json_path – the path to write the JSON

• rel_dir (None) – a relative directory to remove from the filepath of each sample, if possible. The path is converted to an absolute path (if necessary) via fiftyone.core.storage.normalize_path(). The typical use case for this argument is that your source data lives in a single directory and you wish to serialize relative, rather than absolute, paths to the data within that directory

• include_private (False) – whether to include private fields

• include_frames (False) – whether to include the frame labels for video samples

• frame_labels_dir (None) – a directory in which to write per-sample JSON files containing the frame labels for video samples. If omitted, frame labels will be included directly in the returned JSON dict (which can be quite quite large for video datasets containing many frames). Only applicable to datasets that contain videos when include_frames is True

• pretty_print (False) – whether to render the JSON in human readable format with newlines and indentations

aggregate(aggregations, _mongo=False)

Aggregates one or more fiftyone.core.aggregations.Aggregation instances.

Note that it is best practice to group aggregations into a single call to aggregate(), as this will be more efficient than performing multiple aggregations in series.

Parameters:

aggregations – an fiftyone.core.aggregations.Aggregation or iterable of fiftyone.core.aggregations.Aggregation instances

Returns:

Aggregation result(s) corresponding to the input aggregation(s). Returns a single result for a single aggregation, a list of results for multiple aggregations, or a generator for lazy aggregations