Reproduction: From Seeing to Thinking: Decoupling Perception and Reasoning Improves Post-Training of Vision-Language Models