Video Descriptions with Temporal Annotations for 20,000 Videos

The ActivityNet Captions dataset contains 20,000 videos, each annotated with an average of 3.65 temporally localized descriptive sentences, resulting in 100,000 total sentences. Each sentence describes a unique video segment and has an average length of 13.48 words. The dataset was created by Leyo.

Use Cases

• Train a model to generate temporally localized sentence descriptions for unique video segments.
• Analyze the relationship between video duration and the number of annotated sentences per video.
• Build a model to predict sentence length distribution, which averages 13.48 words per description.
• Develop video retrieval systems using the 100,000 sentence descriptions as queries.

Strengths

• Contains 20,000 videos, providing a substantial base for video-language tasks.
• Offers 100,000 temporally annotated sentences, averaging 3.65 sentences per video.
• Sentence descriptions have a defined average length of 13.48 words, indicating consistent annotation detail.

Limitations

• The dataset's geographic and demographic coverage is not specified, which may limit generalizability.
• Annotations are crowdsourced, which can introduce label noise or inconsistency.
• The temporal coverage of the source videos is unknown.

Provenance

Source

Hugging Face (Activitynet Captions).

Collection Method

Videos annotated with temporally localized sentence descriptions, crowdsourced by language creators and generated by experts.

Freshness

Last updated on 2022-07-01.

Geography

Region is listed as 'us', but specific spatial coverage is not detailed.