This directory contains the test dataset used for the CMRAG paper.
All PDFs in this dataset have been rendered and parsed by Qwen2.5-VL-72B:
PDF pages are rendered into images (*.png)
Parsed page content is stored in *_parser.html
Sub-image/region bounding boxes are stored in *_subimg_boxes.json
/ywh/data
├── MMLongBench
│ ├── mmlongbench_retrieval.json
│ └── images/
│ └── <pdf_name>/
│ ├── <page_id>.png
│… See the full description on the dataset page:
https://huggingface.co/datasets/White-white/CMRAG-Bench.
