Vision-R1: Evolving Human-Free Alignment in Large Vision-Language Models via Vision-Guided Reinforcement Learning

models 动态奖励 强化学习 性能提升 视觉引导 视觉语言模型