Reproduce “Xiaomi-OCR-0 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
Xin Chen, Anan Du, Feng Feng, Pei Fu, Jian Luan, Longwei Xu, Shaojie Zhang, Hang Li and 2 more
The authors train Xiaomi-OCR-0 for document parsing and OCR-centric question answering using an approximately 170-million-sample corpus and a progressive recipe that anchors text spatially, continues pretraining across tasks, and applies mixed-task reinforcement learning with task-specific rewards. The 0.8B system scores 95.24 on Real5-OmniDocBench and 83.2 across five OCR-VQA benchmarks, while transfer depends on parsing maturity.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.