MUSAR: Exploring Multi-Subject Customization from Single-Subject Dataset via Attention Routing
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916720185180160 |
|---|---|
| author | Guo, Zinan Zhang, Pengze Wu, Yanze Mou, Chong Zhao, Songtao He, Qian |
| author_facet | Guo, Zinan Zhang, Pengze Wu, Yanze Mou, Chong Zhao, Songtao He, Qian |
| contents | Current multi-subject customization approaches encounter two critical challenges: the difficulty in acquiring diverse multi-subject training data, and attribute entanglement across different subjects. To bridge these gaps, we propose MUSAR - a simple yet effective framework to achieve robust multi-subject customization while requiring only single-subject training data. Firstly, to break the data limitation, we introduce debiased diptych learning. It constructs diptych training pairs from single-subject images to facilitate multi-subject learning, while actively correcting the distribution bias introduced by diptych construction via static attention routing and dual-branch LoRA. Secondly, to eliminate cross-subject entanglement, we introduce dynamic attention routing mechanism, which adaptively establishes bijective mappings between generated images and conditional subjects. This design not only achieves decoupling of multi-subject representations but also maintains scalable generalization performance with increasing reference subjects. Comprehensive experiments demonstrate that our MUSAR outperforms existing methods - even those trained on multi-subject dataset - in image quality, subject consistency, and interaction naturalness, despite requiring only single-subject dataset. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_02823 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | MUSAR: Exploring Multi-Subject Customization from Single-Subject Dataset via Attention Routing Guo, Zinan Zhang, Pengze Wu, Yanze Mou, Chong Zhao, Songtao He, Qian Computer Vision and Pattern Recognition Current multi-subject customization approaches encounter two critical challenges: the difficulty in acquiring diverse multi-subject training data, and attribute entanglement across different subjects. To bridge these gaps, we propose MUSAR - a simple yet effective framework to achieve robust multi-subject customization while requiring only single-subject training data. Firstly, to break the data limitation, we introduce debiased diptych learning. It constructs diptych training pairs from single-subject images to facilitate multi-subject learning, while actively correcting the distribution bias introduced by diptych construction via static attention routing and dual-branch LoRA. Secondly, to eliminate cross-subject entanglement, we introduce dynamic attention routing mechanism, which adaptively establishes bijective mappings between generated images and conditional subjects. This design not only achieves decoupling of multi-subject representations but also maintains scalable generalization performance with increasing reference subjects. Comprehensive experiments demonstrate that our MUSAR outperforms existing methods - even those trained on multi-subject dataset - in image quality, subject consistency, and interaction naturalness, despite requiring only single-subject dataset. |
| title | MUSAR: Exploring Multi-Subject Customization from Single-Subject Dataset via Attention Routing |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2505.02823 |