A Full-Scale Physical Reconstruction Benchmark for Robust Document Parsing in the Wild
Real5-OmniDocBench measures the robustness of document parsing systems under five physical acquisition conditions: Scanning, Warping, Screen-Photography, Illumination, and Skew. It reconstructs the same 1,355 pages from OmniDocBench v1.5 in every condition, producing 6,775 images in total. The one-to-one… See the full description on the dataset page:
https://huggingface.co/datasets/PaddlePaddle/Real5-OmniDocBench.