fiftyone.zoo.datasets.base.ActivityNet100Dataset

ActivityNet is a large-scale video dataset for human activity understanding supporting the tasks of global video classification, trimmed activity classification, and temporal activity detection.

This version contains videos and temporal activity detections for the 100 class version of the dataset.

Notes:

• ActivityNet 100 and 200 differ in the number of activity classes and videos per split
• Partial downloads will download videos (if still available) from YouTube
• Full splits can be loaded by first downloading the official source files from the ActivityNet maintainers
• The test set does not have annotations

Full split stats:

• Train split: 4,819 videos (7,151 instances)
• Test split: 2,480 videos (labels withheld)
• Validation split: 2,383 videos (3,582 instances)

Partial downloads:

• You can specify subsets of data to download via the max_duration, classes, and max_samples parameters

Full split downloads:

Many videos have been removed from YouTube since the creation of ActivityNet. As a result, if you do not specify any partial download parameters described below, you must first download the official source files from the ActivityNet maintainers in order to load a full split into FiftyOne.

To download the source files, you must fill out this form.

Refer to :ref:`this page <activitynet-full-split-downloads>` to see how to load full splits by passing the source_dir parameter to load_zoo_dataset().

Example usage:

import fiftyone as fo
import fiftyone.zoo as foz

#
# Load 10 random samples from the validation split
#
# Only the required videos will be downloaded (if necessary)
#

dataset = foz.load_zoo_dataset(
    "activitynet-100",
    split="validation",
    max_samples=10,
    shuffle=True,
)

session = fo.launch_app(dataset)

#
# Load 10 samples from the validation split that
# contain the actions "Bathing dog" and "Walking the dog"
#
# Videos that contain all `classes` will be prioritized first, followed
# by videos that contain at least one of the required `classes`. If
# there are not enough videos matching `classes` in the split to meet
# `max_samples`, only the available videos will be loaded.
#
# Videos will only be downloaded if necessary
#
# Subsequent partial loads of the validation split will never require
# downloading any videos
#

dataset = foz.load_zoo_dataset(
    "activitynet-100",
    split="validation",
    classes=["Bathing dog", "Walking the dog"],
    max_samples=10,
)

session.dataset = dataset

Dataset size

223 GB

Source

http://activity-net.org/index.html