Problem
Research questions and friction points this paper is trying to address.
Document Parsing
Optical Character Recognition
Vision-Language Models
OCR-centric Understanding
Xiaomi-OCR-0 Technical Report
🤖 AI Summary
This study addresses the reliance of compact OCR models on costly supervision and their limited deep document understanding by proposing a unified 0.8B model based on Qwen3.5. Methodologically, an automated data engine comprising 170 million samples is constructed, leveraging expert consensus and rendering verification to ensure high-quality synthesis. The approach introduces Q-Mask text anchoring alongside a progressive training strategy, and employs mixed-task reinforcement learning (Mix-RL) to jointly optimize document parsing and OCR-centric understanding. Experimental results demonstrate that the proposed model achieves superior performance on benchmarks such as Real5-OmniDocBench, attaining an average score of 83.2 across five OCR VQA tasks. These findings effectively validate that comprehension-oriented supervision significantly enhances parsing capabilities in lightweight architectures.
📝 Abstract
Compact OCR-specific vision-language models achieve strong document parsing performance, but often rely on costly supervision and focus primarily on visual-text reconstruction. We introduce Xiaomi-OCR-0, a unified 0.8B model for document parsing and OCR-centric understanding. We build an approximately 170M-sample OCR-centric corpus using an automated data engine that combines expert consensus, render-based verification, and targeted synthesis. Starting from Qwen3.5-0.8B, our progressive training recipe combines Q-Mask-based text anchoring, continued pretraining, and mixed-task reinforcement learning (Mix-RL). Xiaomi-OCR-0 achieves 95.24 on Real5-OmniDocBench, 96.83 on OmniDocBench v1.6, and 87.94 on Wild-OmniDocBench, while reaching an average score of 83.2 across five OCR-oriented VQA benchmarks. Ablations further show that, with sufficient parsing training, OCR-centric understanding supervision provides additional gains for document parsing.
Homepage: https://huggingface.co/spaces/SeerRay-Lab/Xiaomi-OCR-0.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Compact Vision-Language Model
Automated Data Engine
Progressive Training Recipe
Q-Mask Text Anchoring
Mixed-Task Reinforcement Learning
🔎 Similar Papers
No similar papers found.
