Research map / Vision-language and multimodal / Cross-modal learning
Bilevel Layer-Positioning LoRA for Real Image Dehazing
Yan Zhang, Long Ma, Yuxin Feng, Zhe Huang, Fan Zhou, Zhuo Su
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 positive text guidance to H2C loss (no negative, no positioning) (ablation) | DEA adaptation · Real dehazing average | FADE: 1.018 -> 0.862 | Improved |
| add negative text guidance alone (ablation) | DEA adaptation · Real dehazing average | MUSIQ: 62.05 -> 60.47 | Hurt |
| add both positive and negative text guidance versus positive only (ablation) | DEA adaptation · Real dehazing average | MUSIQ: 62.23 -> 63.31 | Improved |
| add naive (joint) layer positioning to full H2C loss (ablation) | DEA adaptation · Real dehazing average | FADE: 0.774 -> 0.662 | Improved |
| replace naive joint positioning with bilevel positioning under full H2C (ablation) | DEA adaptation · Real dehazing average | FADE: 0.662 -> 0.638 | Improved |
| replace naive joint positioning with bilevel positioning under full H2C (ablation) | DEA adaptation · Real dehazing average | MUSIQ: 64.07 -> 64.40 | Improved |
| add bilevel positioning with positive-only H2C (ablation) | DEA adaptation · Real dehazing average | BIQME: 0.589 -> 0.601 | Improved |
| replace full fine-tuning with BiLaLoRA | DEA adaptation · Full fine-tuning vs BiLaLoRA | Training time (hours): 4.215 -> 0.940 | Improved |
| replace full fine-tuning with BiLaLoRA | DEA adaptation · Full fine-tuning vs BiLaLoRA | FADE: 0.610 -> 0.638 | Hurt |
| replace full fine-tuning with BiLaLoRA | DEA adaptation · Full fine-tuning vs BiLaLoRA | MUSIQ: 64.43 -> 64.40 | No clear effect |
| increase LoRA layers up to three (ablation) | BiLaLoRA · Real dehazing average | Overall quality | Improved |
| apply BiLaLoRA to synthetic-trained DEA | DEA · Real dehazing average | MUSIQ: 57.42 -> 64.40 | Improved |
Closest papers by meaning
- Learning Hazing to Dehazing: Towards Realistic Haze Generation for Real-World Image Dehazing (CVPR 2025 · similarity 0.88)
- Exploiting Diffusion Prior for Real-World Image Dehazing with Unpaired Training (AAAI 2025 · similarity 0.88)
- PromptHaze: Prompting Real-world Dehazing via Depth Anything Model (AAAI 2025 · similarity 0.87)
- Adaptive Dynamic Dehazing via Instruction-Driven and Task-Feedback Closed-Loop Optimization for Diverse Downstream Task Adaptation (AAAI 2026 · similarity 0.87)
- CoA: Towards Real Image Dehazing via Compression-and-Adaptation (CVPR 2025 · similarity 0.87)
- Tokenize Image Patches: Global Context Fusion for Effective Haze Removal in Large Images (CVPR 2025 · 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.
