YouTube Video Frames Spanning 180 Million Images

YT-Temporal-180M is a dataset of 6 million YouTube videos, from which 180 million frames have been extracted. It was created by HuggingFaceM4 and covers diverse topics.

Use Cases

• Train a temporal action recognition model on sequences of 180 million extracted video frames.
• Develop a video topic classification system using the diverse topics covered across 6 million videos.
• Analyze visual content diversity and distribution across the large-scale collection of extracted frames.

Strengths

• Contains 180 million extracted frames, providing a large-scale visual resource.
• Derived from 6 million source videos, ensuring a broad base of content.
• Covers diverse topics, increasing the variety of visual concepts present.

Limitations

• No column or feature-level metadata is provided, limiting structured analysis.
• The dataset's specific video selection criteria and potential biases are unknown.
• The last known update was in May 2022, so the content may not reflect recent videos.

Provenance

Source

HuggingFaceM4

Collection Method

Extracted frames from 6 million YouTube videos.

Freshness

Last updated 2022-05-24.