Request access to the SurgVLM Video Benchmark

This is a held-out evaluation set. Access is granted manually; requests are usually reviewed within a few days.

This dataset is a held-out benchmark. You agree to use it for evaluation only and NOT to include it in any training or fine-tuning data, and not to redistribute it.

Log in or Sign Up to review the conditions and access this dataset content.

SurgVLM Video Benchmark

Held-out video eval set for SurgVLM-style surgical video tasks. Each QA row is paired with an MP4 clip (the listed frames encoded in order).

  • 36 tasks, 200 QA rows each = 7,200 rows
  • 6,512 MP4 clips (rows can share a clip)

This is an evaluation benchmark. Do not train or fine-tune on it.

Contents

Path What
*_bench.jsonl Per-task QA (200 rows each). Rows list image_paths; they do not store video_path.
row_media_map.jsonl Eval join: task + id/line -> media_uid + video_path
media/ MP4 clips + media_index.jsonl (one line per clip)
rationale/ Sidecars for the seven *_rationale tasks

video_path is relative to this repo root, e.g. media/GraSP/GraSP_CASE041_002476-002485_n10.mp4.

Join

task + id  ->  row_media_map.jsonl  ->  video_path  ->  media/*.mp4

Clip identity is media_uid (SHA-256 of {image_paths, clip_duration_sec, fps}). Multiple QA rows can share one clip.

Download

hf download yangd05/SurgVLM-Video-Benchmark --repo-type dataset --local-dir SurgVLM-Video-Benchmark

If you hit Xet Runtime Error / Task Panic, retry with HF_HUB_DISABLE_XET=1.

Downloads last month
8