Back to Home
arXiv AI··Papers & Tech

Improving OCR Faithfulness via Gated and Attenuated On-Policy Distillation

中文摘要

该研究通过门控与衰减在策略蒸馏,动态调整教师指导,提升视觉语言模型的OCR忠实度并防止其错误改写图像文本。

English Summary

This paper introduces Gated and Attenuated On-Policy Distillation to enhance OCR faithfulness in vision-language models, preventing incorrect text rewriting by dynamically adjusting teacher guidance as students improve.

Original Excerpt

arXiv:2609.38282v1 Announce Type: new Abstract: Vision-language models may rewrite anomalous text in images into linguistically plausible expressions, compromising OCR transcription faithfulness. Sequence-level task rewards and local teacher guidance are complementary, but guidance from the same teacher may not remain equally effective as the student improves. Offline analysis shows that supervision from a fixed teacher becomes progressively less favorable as the student improves, both across training checkpoints and across response groups with different task rewards. Motivated by this observation, we introduce GAD-RL, which adaptively regulates teacher supervision during joint post-training according to the student's current task performance and local distributions. A frozen teacher conditions on reference transcriptions and student-generated prefixes. GAD-RL disables distillation for response groups containing an output with task reward at least 0.95 and continuously attenuates distillation strength as group-mean reward increases. It also weights forward KL by the student's probability of the teacher's Top-1 token, moderating local auxiliary updates when student support for that …