Reading or Reasoning? Format Decoupled Reinforcement Learning for Document OCR

TL;DR

This paper introduces format decoupled reinforcement learning (FD-RL) to improve OCR performance on formatted documents by leveraging entropy patterns and format-specific rewards, achieving state-of-the-art results on OmniDocBench.

Contribution

The paper proposes a novel FD-RL approach that uses entropy-based filtering and format-specific rewards to enhance OCR accuracy on complex, format-sensitive documents.

Findings

FD-RL achieves 90.41 on OmniDocBench, setting a new record.

Entropy patterns reveal OCR struggles with formatted text.

Ablation studies validate the effectiveness of the proposed strategies.

Abstract

Reading text from images or scanned documents via OCR models has been a longstanding focus of researchers. Intuitively, text reading is perceived as a straightforward perceptual task, and existing work primarily focuses on constructing enriched data engineering to enhance SFT capabilities. In this work, we observe that even advanced OCR models exhibit significantly higher entropy in formatted text (\emph{e.g.}, formula, table, etc.) compared to plain text, often by an order of magnitude. These statistical patterns reveal that advanced OCR models struggle with high output uncertainty when dealing with format sensitive document, suggesting that reasoning over diverse reading pathways may improve OCR performance. To address this, we propose format decoupled reinforcement learning (FD-RL), which leverages high-entropy patterns for targeted optimization. Our approach employs entropy-based data filtration strategy to identify format-intensive instances, and adopt format decoupled rewards tailored to different format types, enabling format-level validation rather than token-level memorization. FD-RL achieves an average score of 90.41 on OmniDocBench, setting a new record for end-to-end models on this highly popular benchmark. More importantly, we conduct comprehensive ablation studies over data, training, filtering, and rewarding strategies, thoroughly validating their effectiveness.