Reproduce “OvisOCR3 Technical Report” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yu Xia, Yinglun Li, Yuhui Chen, Jinlong Huang, Yuyang Wanyan, Shengze Shi, Yifei Song, Yue Duan and 7 more
OvisOCR3 handles document parsing, general OCR, key information extraction, figure captioning, and cross-page table reconstruction with one 0.8B-parameter model. Its training combines domain-balanced loss normalization, distillation from five task-specific teachers, and mixed-task reinforcement learning. On English OCRBench v2, it scores 65.9 versus 54.1 for Qwen3.5-4B, using one-fifth as many parameters; on CrossTabBench, it leads in continuation detection but trails in reconstruction TEDS.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.