通过门控与衰减的在线策略蒸馏提升OCR忠实度
原标题：Improving OCR Faithfulness via Gated and Attenuated On-Policy Distillation
AI 摘要
该论文针对视觉语言模型在OCR中因语言先验而改写图像中异常文本、损害转录忠实度的问题，提出GAD-RL方法。作者发现固定教师模型的监督信号会随学生模型能力提升而逐渐失效，因此设计组级门控与奖励感知衰减机制，在联合后训练中动态调节教师干预强度。在Qwen3.5-2B上，GAD-RL在CHAOS-Bench的扰动词Micro Recall达到59.92%，比GRPO和固定权重GRPO+OPD分别高出8.45和4.43个百分点，同时在OmniDocBench v1.6上保持91.18的总体得分。
正文节选
Improving OCR Faithfulness via Gated and Attenuated On-Policy Distillation 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