Dr. Seg：通过面向感知的设计重新审视视觉大语言模型的 GRPO 训练
Dr. Seg: Revisiting GRPO Training for Visual Large Language Models through Perception-Oriented Design
针对视觉感知任务（如推理分割）重新设计 GRPO 训练框架，提出 Look-to-Confirm 机制和 Distribution-Ranked Reward 模块，无需修改模型架构即可提升 VLLM 在复杂视觉场景下的感知性能，为将强化学习应用于视觉任务提供了可复用的训练策略。
原文依据 we propose Dr.Seg, a simple, plug-and-play GRPO-based framework consisting of a Look-to-Confirm mechanism and a Distribution-Ranked Reward module
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论随着 Group Relative Policy Optimization (GRPO) 在基础大语言模型中的成功，越来越多的研究试图将 GRPO 适配到视觉大语言模型（VLLM）以用于视觉感知任务（例如检测和分割）。然而，这一研究方向中的许多工作都基于一个长期存在但未经检验的假设：为语言推理开发的训练范式可以无缝迁移到视觉感知。我们的实验表明这一假设并不成立，揭示了面向推理与面向感知的设置之间的内在差异。以推理分割作为代表性案例，我们揭示了两个被忽视的因素：(i) 需要更广泛的输出空间，以及 (ii) 细粒度、稳定奖励的重要性。基于这些观察，我们提出了 Dr.Seg，一个简单、即插即用的基于 GRPO 的框架，由 Look-to-Confirm 机制和 Distribution-Ranked Reward 模块组成，无需架构修改，并能与现有的基于 GRPO 的 VLLM 无缝集成。大量实验表明，Dr.Seg 在复杂视觉场景中提升了性能，同时保持了强大的泛化能力。代码、数据和模型将在接收后公开。
英文原文摘要 1259 字符
Following the success of Group Relative Policy Optimization (GRPO) in foundation LLMs, an increasing number of works have sought to adapt GRPO to Visual Large Language Models (VLLMs) for visual perception tasks (e.g., detection and segmentation). However, much of this line of research rests on a long-standing yet unexamined assumption: training paradigms developed for language reasoning can be transferred seamlessly to visual perception. Our experiments show that this assumption is not valid, revealing intrinsic differences between reasoning-oriented and perception-oriented settings. Using reasoning segmentation as a representative case, we surface two overlooked factors: (i) the need for a broader output space, and (ii) the importance of fine-grained, stable rewards. Building on these observations, we propose Dr.Seg, a simple, plug-and-play GRPO-based framework consisting of a Look-to-Confirm mechanism and a Distribution-Ranked Reward module, requiring no architectural modifications and integrating seamlessly with existing GRPO-based VLLMs. Extensive experiments demonstrate that Dr.Seg improves performance in complex visual scenarios while maintaining strong generalization. Code, data, and models will be publicly released upon acceptance.