CRAFT-MAGMaR: Multimodal Video Dataset with ASR and Chunking

CRAFT-MAGMaR combines the MAGMaR-2026 test release with CRAFT pipeline artefacts. The dataset includes 90 original videos and 72 CRAFT-produced video chunks, along with associated automatic speech recognition outputs. It was created by mbhosale and last updated on Hugging Face in May 2026.

Use Cases

• Training video chunking models based on the adaptive keyframe selection process described.
• Evaluating multilingual automatic speech recognition systems on the provided ASR outputs.
• Developing multimodal embeddings using the combined video and ASR data.
• Benchmarking end-to-end video processing pipelines that require synchronized video and text data.

Strengths

• Contains 90 original video files.
• Includes 72 additional video chunks produced by the CRAFT pipeline.
• Provides multilingual automatic speech recognition outputs.

Limitations

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

Provenance

Source

akhilvssg/magmar-2026-test-asr-embeddings and CRAFT pipeline.

Collection Method

Combines a test release with content produced by a pipeline involving chunking, multilingual ASR, and adaptive keyframe selection.

Freshness

Last updated 2026-05-25 00:35:47; freshness should be verified.