Getting Started with FiftyOne Enterprise

Follow this guide to create your first dataset in FiftyOne Enterprise 🚀

Configure cloud credentials

An admin user must configure cloud credentials once for a deployment in order for users to view datasets:

Create a dataset via the SDK

Install the FiftyOne Enterprise Python SDK

1. Navigate to the Settings > API keys page

2. Copy and execute the provided bash command to install the SDK in your virtual environment

If you plan to work with video datasets, you’ll also need to install FFmpeg:

LinuxmacOSWindows

sudo apt install -y ffmpeg

Connect to your deployment

To connect to your FiftyOne Enterprise deployment, you must provide your API URI and API key:

export FIFTYONE_API_URI=XXXXXXXX
export FIFTYONE_API_KEY=YYYYYYYY

You can create an API key and locate your deployment’s URI on the Settings > API keys page of the FiftyOne Enterprise App:

You can use the fiftyone config CLI method to verify that you have correctly configured your API URI and API key:

$ fiftyone config
{
    ...
    "api_uri": "XXXXXXXX",
    "api_key": "YYYYYYYY",
    ...
}

You can also verify that your API connection is working correctly by executing the following method:

# if this fails, you may have the open source SDK installed
import fiftyone.management as fom

# if this succeeds, your API connection is working
fom.test_api_connection()

Set cloud credentials locally

Next, configure the appropriate environment variables to register your cloud credentials in your local environment:

AWSGCPAzureMinIO

Set the following environment variables:

export AWS_ACCESS_KEY_ID=...
export AWS_SECRET_ACCESS_KEY=...
export AWS_DEFAULT_REGION=...

Refer to this page for more information about interacting with cloud-backed media in FiftyOne Enterprise.

Import your data

The example code below shows the basic pattern for creating new datasets and populating them via the FiftyOne Enterprise Python SDK:

AWSGCPAzureMinIO

import fiftyone as fo
import fiftyone.core.storage as fos

dataset = fo.Dataset("<name>")

s3_files = fos.list_files("s3://<bucket>/<prefix>", abs_paths=True)

samples = []
for s3_uri in s3_files:
    if s3_uri.lower().endswith(".jpeg"):
        sample = fo.Sample(filepath=s3_uri)
        samples.append(sample)

dataset.add_samples(samples)

# You must mark the dataset as persistent to access it in the UI
dataset.persistent = True

Refer to this page for more information about importing your media and labels into FiftyOne via Python.

Compute metadata

All datasets/views provide a builtin compute_metadata() method that you can invoke to efficiently populate the metadata field of your samples with basic media type-specific metadata such as file size and image/video dimensions for all samples in a collection:

dataset.compute_metadata()

sample = dataset.first()
print(sample.metadata)

It is highly recommended to keep the metadata field populated for all samples of your datasets because it provides useful information upon which to search/filter and it enables the sample grid’s tiling algorithm to run more efficiently.

You can verify that all samples in a dataset/view have metadata as follows:

assert len(dataset.exists("metadata", False)) == 0

Create a dataset via the UI

Note

An admin must follow these instructions to install the @voxel51/io and @voxel51/utils plugins in order for users to perform imports and compute metadata via the FiftyOne Enterprise UI.

Import your data

To create a new dataset, click on the “New dataset” button in the upper right corner of the FiftyOne Enterprise homepage. A pop-up will appear alowing you to choose a name and optional description/tags for the dataset:

You can then use the import_samples operator to import media and labels stored in a cloud storage bucket:

Compute metadata

You can use the compute_metadata operator to efficiently populate the metadata field of your samples with basic media type-specific metadata such as file size and image/video dimensions for all samples in a collection:

It is highly recommended to keep the metadata field populated for all samples of your datasets because it provides useful information upon which to search/filter and it enables the sample grid’s tiling algorithm to run more efficiently.