From Sparse to Dense: Multi-View GRPO for Flow Models via Augmented Condition Space
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866911511948034048 |
|---|---|
| author | Bu, Jiazi Ling, Pengyang Zhou, Yujie Wang, Yibin Zang, Yuhang Wei, Tianyi Zhan, Xiaohang Wang, Jiaqi Wu, Tong Pan, Xingang Lin, Dahua |
| author_facet | Bu, Jiazi Ling, Pengyang Zhou, Yujie Wang, Yibin Zang, Yuhang Wei, Tianyi Zhan, Xiaohang Wang, Jiaqi Wu, Tong Pan, Xingang Lin, Dahua |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_12648 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | From Sparse to Dense: Multi-View GRPO for Flow Models via Augmented Condition Space Bu, Jiazi Ling, Pengyang Zhou, Yujie Wang, Yibin Zang, Yuhang Wei, Tianyi Zhan, Xiaohang Wang, Jiaqi Wu, Tong Pan, Xingang Lin, Dahua Computer Vision and Pattern Recognition 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. |
| title | From Sparse to Dense: Multi-View GRPO for Flow Models via Augmented Condition Space |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2603.12648 |