fiftyone.utils.tracking.deepsort#

DeepSort wrapper for FiftyOne.

Copyright 2017-2025, Voxel51, Inc.
voxel51.com

Classes:

DeepSort()

class fiftyone.utils.tracking.deepsort.DeepSort#

Bases: object

Methods:

track(sample_collection, in_field[, ...])

Performs object tracking using the DeepSort algorithm on the given video samples.

track_sample(sample, in_field[, out_field, ...])

Performs object tracking using the DeepSort algorithm on the given video sample.
static track(sample_collection, in_field, out_field='frames.ds_tracks', max_age=5, keep_confidence=False, skip_failures=True, progress=None)#

Performs object tracking using the DeepSort algorithm on the given video samples.

DeepSort is an algorithm for tracking multiple objects in video streams based on deep learning techniques. It associates bounding boxes between frames and maintains tracks of objects over time.

Parameters:

  • sample_collection – a fiftyone.core.collections.SampleCollection

  • in_field – the name of a frame field containing fiftyone.core.labels.Detections to track. The "frames." prefix is optional

  • out_field ("frames.ds_tracks") – the name of a frame field to store the output fiftyone.core.labels.Detections with tracking information. The "frames." prefix is optional

  • max_age (5) – the maximum number of missed misses before a track is deleted

  • keep_confidence (False) – whether to store the detection confidence of the tracked objects in the out_field

  • skip_failures (True) – whether to gracefully continue without raising an error if tracking fails for a video

  • 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

static track_sample(sample, in_field, out_field='ds_tracks', max_age=5, keep_confidence=False)#

Performs object tracking using the DeepSort algorithm on the given video sample.

DeepSort is an algorithm for tracking multiple objects in video streams based on deep learning techniques. It associates bounding boxes between frames and maintains tracks of objects over time.

Parameters:

  • sample – a fiftyone.core.sample.Sample

  • in_field – the name of the frame field containing fiftyone.core.labels.Detections to track

  • out_field ("ds_tracks") – the name of a frame field to store the output fiftyone.core.labels.Detections with tracking information. The "frames." prefix is optional

  • max_age (5) – the maximum number of missed misses before a track is deleted

  • keep_confidence (False) – whether to store the detection confidence of the tracked objects in the out_field