研究人员推出CycleGRPO，一个新颖的强化学习框架，旨在统一多模态大语言模型（MLLMs）的区域理解与定位。该框架在一个自我评估范式下运行，其中MLLM既充当生成区域标题的Actor，又充当将这些标题重新映射回空间域的Critic。CycleGRPO仅使用区域输入和质量感知的循环一致性奖励，绕过了对文本真实值（ground truth）的需求，并在各种基准测试中展示了性能提升，且无需针对特定任务进行微调。 AI
影响 该框架有望提升MLLMs的像素级能力，可能提高它们在区域标题生成和指代分割等任务中的性能。
排序理由 该集群描述了一篇详细介绍多模态大语言模型新框架的研究论文。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2607.11581v1 Announce Type: new Abstract: This paper introduces Actor as Its Own Critic, a unified reinforcement learning framework, Cycle Group Relative Policy Optimization (CycleGRPO), that jointly optimizes region understanding and localization for Multimodal Large Langu…
This paper introduces Actor as Its Own Critic, a unified reinforcement learning framework, Cycle Group Relative Policy Optimization (CycleGRPO), that jointly optimizes region understanding and localization for Multimodal Large Language Models (MLLMs). Unlike existing separate pip…