arXiv：cs.AI· Wenjie Liao, Xiaohui Song, Liangjie Zhao, Haonan Lu·· 3 小时前
SP-DocReader：面向精准文档 OCR 的差异感知自博弈框架
SP-DocReader: Difference-Aware Self-Play for Precise Document OCR
AI 导读
SP-DocReader 是一个针对监督微调后残余误差的 OCR 自博弈框架，提出 Reading Discrepancy Masking 与 Focused Fidelity Loss，仅训练 OCR 模块、冻结主干。
正文
Abstract:Accurate page transcription remains difficult for vision language models under limited input and training budgets. We present SP-DocReader, a self-play framework for optical character recognition (OCR) that targets residual errors after supervised fine-tuning. Reading Discrepancy Masking aligns reference and generated model tokens through a longest common subsequence, then scores unmatched positions with their full conditioning prefixes. Focused Fidelity Loss adds direct negative log-likelihood supervision at unmatched ground-truth positions. Only the OCR module is trained, while the backbone remains frozen. We derive the combined gradient to distinguish relative score optimization from direct supervision. Compared with SFT-2, SP-DR-3 reduces Vary-600K character error rate on both backbones. On Qwen3-VL-4B, it reduces character error rate by approximately 54 percent and improves DocVQA Average Normalized Levenshtein Similarity (ANLS) by 3.7 points. These results show the value of focusing self-play training on the discrepancies that remain after supervised fine-tuning.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.11148 [cs.CV] |
| (or arXiv:2610.11148v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2610.11148 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Wenjie Liao [view email]
[v1]
Thu, 8 Oct 2026 03:12:12 UTC (1,030 KB)
来源：arXiv：cs.AI · arxiv.org