Search papers, labs, and topics across Lattice.
Multi-View GRPO (MV-GRPO) is introduced to improve preference alignment in text-to-image flow models by addressing the limitations of single-view evaluation in Group Relative Policy Optimization (GRPO). MV-GRPO augments the condition space with semantically adjacent captions generated by a Condition Enhancer, enabling multi-view advantage re-estimation and richer optimization signals. Experiments show MV-GRPO achieves superior alignment performance compared to existing methods by leveraging the probability distribution of original samples conditioned on the new captions without sample regeneration.
Text-to-image flow models get a preference alignment boost by generating multiple related captions per image, creating a richer reward landscape without expensive re-sampling.
Group Relative Policy Optimization (GRPO) has emerged as a powerful framework for preference alignment in text-to-image (T2I) flow models. However, we observe that the standard paradigm where evaluating a group of generated samples against a single condition suffers from insufficient exploration of inter-sample relationships, constraining both alignment efficacy and performance ceilings. To address this sparse single-view evaluation scheme, we propose Multi-View GRPO (MV-GRPO), a novel approach that enhances relationship exploration by augmenting the condition space to create a dense multi-view reward mapping. Specifically, for a group of samples generated from one prompt, MV-GRPO leverages a flexible Condition Enhancer to generate semantically adjacent yet diverse captions. These captions enable multi-view advantage re-estimation, capturing diverse semantic attributes and providing richer optimization signals. By deriving the probability distribution of the original samples conditioned on these new captions, we can incorporate them into the training process without costly sample regeneration. Extensive experiments demonstrate that MV-GRPO achieves superior alignment performance over state-of-the-art methods.
