Chapter Llama: VidChapters-7M Dataset for Video Chaptering

VidChapters-7M contains 817,000 videos with associated Automatic Speech Recognition data and chapter annotations. The dataset was created by lucas-ventura for the CVPR 2025 paper on efficient chaptering in hour-long videos. It includes captions extracted using various sampling strategies and chapter titles with timestamps.

Use Cases

• Training video chaptering models based on ASR transcripts and timestamped annotations.
• Developing multimodal models that align visual content with spoken language for segmentation.
• Benchmarking automatic chapter title generation algorithms based on video and audio content.

Strengths

• Contains 817,000 videos, indicating a large-scale collection.
• Includes both ASR data and chapter annotations with timestamps and titles.
• Associated with a peer-reviewed CVPR 2025 publication.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the ASR and chapter data is unknown, which may limit suitability assessment.
• The description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

lucas-ventura

Collection Method

Captions extracted from videos using various sampling strategies.

Freshness

Last updated 2025-10-20 09:31:44