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.
Keywords
Subject
Publication details
- Journal
- Not available
- Open access
- Green open access
Cite this article
APA 7
Liao, W., Song, X., Zhao, L., & Lu, H. (2026). SP-DocReader: Difference-Aware Self-Play for Precise Document OCR. https://omanscience.com/en/articles/sp-docreader-difference-aware-self-play-for-precise-document-ocr
MLA 9
Liao, Wenjie, et al. "SP-DocReader: Difference-Aware Self-Play for Precise Document OCR." https://omanscience.com/en/articles/sp-docreader-difference-aware-self-play-for-precise-document-ocr.
Chicago (author–date)
Liao, Wenjie, Xiaohui Song, Liangjie Zhao, and Haonan Lu. 2026. "SP-DocReader: Difference-Aware Self-Play for Precise Document OCR." https://omanscience.com/en/articles/sp-docreader-difference-aware-self-play-for-precise-document-ocr.
Harvard
Liao, W., Song, X., Zhao, L. and Lu, H. (2026) 'SP-DocReader: Difference-Aware Self-Play for Precise Document OCR', Available at: https://omanscience.com/en/articles/sp-docreader-difference-aware-self-play-for-precise-document-ocr.
Vancouver
Liao W, Song X, Zhao L, Lu H. SP-DocReader: Difference-Aware Self-Play for Precise Document OCR. https://omanscience.com/en/articles/sp-docreader-difference-aware-self-play-for-precise-document-ocr
IEEE
W. Liao, X. Song, L. Zhao, and H. Lu, "SP-DocReader: Difference-Aware Self-Play for Precise Document OCR," https://omanscience.com/en/articles/sp-docreader-difference-aware-self-play-for-precise-document-ocr.