VLM-R^3: Region Recognition, Reasoning, and Refinement for Enhanced Multimodal Chain-of-Thought Paper • 2505.16192 • Published May 22 • 12