让AI先看图再推理，避免依赖文字提示的捷径
Seeing Before Reasoning: Decoupling Perception and Reasoning for Shortcut-Resilient Multimodal On-Policy Self-Distillation
- 分两步训练：先描述图像，再基于描述推理答案
- 图像教师监督描述，推理教师监督答案，提升视觉对齐
- 适合需要准确理解图像的多模态任务，如视觉数学与空间定位
在策略自蒸馏（OPSD）中，模型通过自身生成的轨迹进行训练，并利用一个冻结的副本提供基于参考目标的密集词级监督。该方法在大语言模型推理中表现良好，但直接应用于多模态大语言模型（MLLM）时可能产生捷径：特权目标可能主要依赖文本参考而非图像。为此，我们提出ViGOS，一种用于MLLM后训练的视觉引导型OPSD框架。学生模型首先生成视觉描述，再据此推理最终答案。对于有效轨迹，仅图像的感知教师监督描述部分，而特权推理教师监督推理过程和最终答案；对于无效轨迹，则使用参考教师恢复输出格式。在通用视觉-语言、专家推理、视觉数学、空间定位及视觉-语言先验等多个基准上，ViGOS保持了OPSD的主要优势，并显著提升了在易产生捷径场景下的图像对齐行为。
原文摘要 · Abstract (English)
On-policy self-distillation (OPSD) trains a model on its own rollouts and uses a frozen copy to provide dense token-level targets conditioned on a reference target. This works well for LLM reasoning, but a direct extension to multimodal large language models (MLLMs) can create a shortcut: the privileged target may guide tokens mainly based on the text reference target rather than the image. We propose ViGOS, a visually grounded OPSD framework for MLLM post-training. The student first writes a visual description and then reasons toward the final answer. For valid rollouts, an image-only perception teacher supervises the description, while a privileged reasoning teacher supervises the reasoning and final answer on the same student prefix. A reference teacher is used only for invalid rollouts to recover the output format. Across general vision-language, expert reasoning, visual math, spatial grounding, and visual-language-prior benchmarks, ViGOS keeps the main benefits of OPSD and improves image-grounded behavior in shortcut-prone settings.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。