VirtualLUO/

Wild_OmniDocBench

A Real-World Captured Document Parsing Benchmark for Robustness Evaluation

中文版 • Paper • GitHub • HuggingFace

Wild-OmniDocBench is a benchmark for evaluating document parsing robustness under real-world captured conditions. It is derived from OmniDocBench by converting scanned/digital documents into naturally captured images through controlled physical simulation, including printing, deformation, and photography under diverse… See the full description on the dataset page: https://huggingface.co/datasets/VirtualLUO/Wild_OmniDocBench.