This is a self-contained, 2,298-example multiple-choice surgical VQA validation
benchmark derived from CholecT45 test videos 74, 75, 78, 79, and 80. It has six
subsets: L1 (504), L2 (504), L3 (390), L4 (300), CVS (300), and DYN (300).
Dataset structure
The dataset is Parquet-sharded for Hub preview and streaming. image contains the
primary PNG frame as bytes; image_after contains the second frame for DYN examples
and is null… See the full description on the dataset page: https://huggingface.co/datasets/EgoF0102/cholec-scene-graph-val.