Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2604.25636 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911628974358528 |
|---|---|
| author | Guo, Jiayi Wang, Linqing Wang, Jiangshan Yue, Yang Liu, Zeyu Zhao, Zhiyuan Lu, Qinglin Huang, Gao Wang, Chunyu |
| author_facet | Guo, Jiayi Wang, Linqing Wang, Jiangshan Yue, Yang Liu, Zeyu Zhao, Zhiyuan Lu, Qinglin Huang, Gao Wang, Chunyu |
| contents | Unified multimodal models (UMMs) integrate visual understanding and generation within a single framework. For text-to-image (T2I) tasks, this unified capability allows UMMs to refine outputs after their initial generation, potentially extending the performance upper bound. Current UMM-based refinement methods primarily follow a refinement-via-editing (RvE) paradigm, where UMMs produce editing instructions to modify misaligned regions while preserving aligned content. However, editing instructions often describe prompt-image misalignment only coarsely, leading to incomplete refinement. Moreover, pixel-level preservation, though necessary for editing, unnecessarily restricts the effective modification space for refinement. To address these limitations, we propose Refinement via Regeneration (RvR), a novel framework that reformulates refinement as conditional image regeneration rather than editing. Instead of relying on editing instructions and enforcing strict content preservation, RvR regenerates images conditioned on the target prompt and the semantic tokens of the initial image, enabling more complete semantic alignment with a larger modification space. Extensive experiments demonstrate the effectiveness of RvR, improving Geneval from 0.78 to 0.91, DPGBench from 84.02 to 87.21, and UniGenBench++ from 61.53 to 77.41. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_25636 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Refinement via Regeneration: Enlarging Modification Space Boosts Image Refinement in Unified Multimodal Models Guo, Jiayi Wang, Linqing Wang, Jiangshan Yue, Yang Liu, Zeyu Zhao, Zhiyuan Lu, Qinglin Huang, Gao Wang, Chunyu Computer Vision and Pattern Recognition Unified multimodal models (UMMs) integrate visual understanding and generation within a single framework. For text-to-image (T2I) tasks, this unified capability allows UMMs to refine outputs after their initial generation, potentially extending the performance upper bound. Current UMM-based refinement methods primarily follow a refinement-via-editing (RvE) paradigm, where UMMs produce editing instructions to modify misaligned regions while preserving aligned content. However, editing instructions often describe prompt-image misalignment only coarsely, leading to incomplete refinement. Moreover, pixel-level preservation, though necessary for editing, unnecessarily restricts the effective modification space for refinement. To address these limitations, we propose Refinement via Regeneration (RvR), a novel framework that reformulates refinement as conditional image regeneration rather than editing. Instead of relying on editing instructions and enforcing strict content preservation, RvR regenerates images conditioned on the target prompt and the semantic tokens of the initial image, enabling more complete semantic alignment with a larger modification space. Extensive experiments demonstrate the effectiveness of RvR, improving Geneval from 0.78 to 0.91, DPGBench from 84.02 to 87.21, and UniGenBench++ from 61.53 to 77.41. |
| title | Refinement via Regeneration: Enlarging Modification Space Boosts Image Refinement in Unified Multimodal Models |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2604.25636 |