Note for reviewers: every JSON file in this repo is a test-only
sub-task. We declare three configs (2D, 3D, Text) and route all
files to the test split explicitly so the viewer does not mis-group
them based on filename heuristics. Image fields (qry_img_path,
tgt_img_path) are relative paths into the matching 2D_Images/ /
3D_Images/ folders.
This repository is a small, self-contained sample of the full MedHEB-Bench
benchmark, hosted here so… See the full description on the dataset page:
https://huggingface.co/datasets/MEDHEB/MedHEB-Bench_Sample.
