Research map / Vision-language and multimodal / Multimodal LLMs
Towards Real-World Adverse Weather Image Restoration: Enhancing Clearness and Semantics with Vision-Language Models
Jiaqi Xu, Mengyang Wu, Xiaowei Hu, Chi-Wing Fu, Qi Dou, Pheng-Ann Heng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add mean-teacher pseudo-labels from unlabeled real images (ablation) | supervised MSBDN · real adverse weather overall | MUSIQ: 53.41 -> 54.08 | Improved |
| add VLM-based image assessment to select pseudo-labels (ablation) | semi-supervised MSBDN · real adverse weather overall | MUSIQ: 54.08 -> 56.68 | Improved |
| add pseudo-label initialization from existing restoration methods (ablation) | semi-supervised MSBDN · real adverse weather overall | MUSIQ: 56.68 -> 57.34 | Improved |
| add pseudo-label initialization (ablation) | semi-supervised MSBDN · real adverse weather overall | CLIP-IQA: 0.429 -> 0.425 | No clear effect |
| add weather prompt learning (CLIP prompts for clear vs weather) (ablation) | semi-supervised MSBDN · real adverse weather overall | MUSIQ: 57.34 -> 58.13 | Improved |
| add description-assisted semantics regularization (ablation) | semi-supervised MSBDN · real adverse weather overall | MUSIQ: 58.13 -> 58.91 | Improved |
| add iterative update of pseudo-labels (ablation) | semi-supervised MSBDN · real adverse weather overall | MUSIQ: 58.91 -> 59.34 | Improved |
| select pseudo-labels with VLM rating instead of NIMA, MUSIQ, CLIP-IQA or LIQE (ablation) | semi-supervised MSBDN · real adverse weather | visual quality of restoration | Improved |
| train with unlabeled real images and VLM guidance versus synthetic-only restoration baselines | adverse weather image restoration · real adverse weather overall | MUSIQ: 56.38 -> 59.34 | Improved |
| use proposed method versus PromptIR on real haze images | adverse weather image restoration · real haze | CLIP-IQA: 0.372 vs 0.371 | No clear effect |
Closest papers by meaning
- Language-driven All-in-one Adverse Weather Removal (CVPR 2024 · similarity 0.90)
- Real-World Adverse Weather Image Restoration via Dual-Level Reinforcement Learning with High-Quality Cold Start (NeurIPS 2025 · similarity 0.90)
- AWRaCLe: All-Weather Image Restoration Using Visual In-Context Learning (AAAI 2025 · similarity 0.89)
- Continuous Adverse Weather Removal via Degradation-Aware Distillation (CVPR 2025 · similarity 0.88)
- Semantic Segmentation in Multiple Adverse Weather Conditions with Domain Knowledge Retention (AAAI 2024 · similarity 0.87)
- Learning Real-World Image De-weathering with Imperfect Supervision (AAAI 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
