Zero-Reference Joint Low-Light Enhancement and Deblurring via Visual Autoregressive Modeling with VLM-Derived Modulation
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_ | 1866911283410894848 |
|---|---|
| author | Dong, Wei Zhou, Han Lin, Junwei Chen, Jun |
| author_facet | Dong, Wei Zhou, Han Lin, Junwei Chen, Jun |
| contents | Real-world dark images commonly exhibit not only low visibility and contrast but also complex noise and blur, posing significant restoration challenges. Existing methods often rely on paired data or fail to model dynamic illumination and blur characteristics, leading to poor generalization. To tackle this, we propose a generative framework based on visual autoregressive (VAR) modeling, guided by perceptual priors from the vision-language model (VLM). Specifically, to supply informative conditioning cues for VAR models, we deploy an adaptive curve estimation scheme to modulate the diverse illumination based on VLM-derived visibility scores. In addition, we integrate dynamic and spatial-frequency-aware Rotary Positional Encodings (SF-RoPE) into VAR to enhance its ability to model structures degraded by blur. Furthermore, we propose a recursive phase-domain modulation strategy that mitigates blur-induced artifacts in the phase domain via bounded iterative refinement guided by VLM-assessed blur scores. Our framework is fully unsupervised and achieves state-of-the-art performance on benchmark datasets. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_18591 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Zero-Reference Joint Low-Light Enhancement and Deblurring via Visual Autoregressive Modeling with VLM-Derived Modulation Dong, Wei Zhou, Han Lin, Junwei Chen, Jun Computer Vision and Pattern Recognition Real-world dark images commonly exhibit not only low visibility and contrast but also complex noise and blur, posing significant restoration challenges. Existing methods often rely on paired data or fail to model dynamic illumination and blur characteristics, leading to poor generalization. To tackle this, we propose a generative framework based on visual autoregressive (VAR) modeling, guided by perceptual priors from the vision-language model (VLM). Specifically, to supply informative conditioning cues for VAR models, we deploy an adaptive curve estimation scheme to modulate the diverse illumination based on VLM-derived visibility scores. In addition, we integrate dynamic and spatial-frequency-aware Rotary Positional Encodings (SF-RoPE) into VAR to enhance its ability to model structures degraded by blur. Furthermore, we propose a recursive phase-domain modulation strategy that mitigates blur-induced artifacts in the phase domain via bounded iterative refinement guided by VLM-assessed blur scores. Our framework is fully unsupervised and achieves state-of-the-art performance on benchmark datasets. |
| title | Zero-Reference Joint Low-Light Enhancement and Deblurring via Visual Autoregressive Modeling with VLM-Derived Modulation |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2511.18591 |