1 paper · 1 filter
Junzhe Li, Yutao Cui, Tao Huang +8
Although GRPO substantially enhances flow matching models in human preference alignment of image generation, methods such as FlowGRPO and DanceGRPO still exhibit inefficiency due t…