OvisOCR2 Technical Report
A 0.8B end-to-end document parser tops OmniDocBench v1.6 with a 96.58 overall score.
OvisOCR2 turns a document page image into Markdown in natural reading order, including text, formulas, tables, and visual regions. The authors train it with filtered real-document annotations, synthetic HTML-derived pages, supervised fine-tuning, reinforcement learning on a 4B branch, on-policy distillation, and model fusion. They also report the highest Avg3 score on PureDocBench, at 75.06. An in-house benchmark is cited as further evidence of performance on harder long-tail cases. HF Daily Papers' note
OvisOCR2 turns a document page image into Markdown in natural reading order, including text, formulas, tables, and visual regions. The authors train it with filtered real-document annotations, synthetic HTML-derived pages, supervised fine-tuning, reinforcement learning on a 4B branch, on-policy distillation, and model fusion. They also report the highest Avg3 score on PureDocBench, at 75.06. An in-house benchmark is cited as further evidence of performance on harder long-tail cases. HF Daily Papers' note
score 6