Research map / Reinforcement learning / Visual and representation RL
Physics Meets Perception: A Reinforcement Learning Framework for Unpaired Real-World Image Dehazing
Yunwei Lan, Zhigao Cui, Chang Liu, Menglin Zhang, Nian Wang, Cong Zhang, Dong Liu
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 GRPO RL fine-tuning after GAN warmup (ablation) | Dehaze-RL · RTTS | MUSIQ: 58.927 -> 63.921 | Improved |
| add GRPO RL fine-tuning after GAN warmup (ablation) | Dehaze-RL · RTTS | NIQE: 4.226 -> 3.641 | Improved |
| add GRPO RL fine-tuning after supervised L1 warmup (ablation) | Dehaze-RL · RTTS (synthetic-warmup) | MUSIQ: 53.667 -> 55.994 | Improved |
| add GRPO RL fine-tuning after supervised L1 warmup (ablation) | Dehaze-RL · RTTS (synthetic-warmup) | CLIPIQA: 0.304 -> 0.355 | Improved |
| add semantic CLIP reward (ablation) | Dehaze-RL · RTTS | MUSIQ: 62.212 -> 63.921 | Improved |
| add no-reference perceptual metric rewards (MUSIQ, NIQE) (ablation) | Dehaze-RL · RTTS | MUSIQ: 61.575 -> 63.921 | Improved |
| add VLM feedback reward (ablation) | Dehaze-RL · RTTS | NIQE: 3.776 -> 3.641 | Improved |
| add structural reward (PatchNCE and physical reversibility) (ablation) | Dehaze-RL · RTTS | MANIQA: 0.159 -> 0.168 | Improved |
| add structural reward (ablation) | Dehaze-RL · RTTS | CLIPIQA: 0.425 -> 0.427 | No clear effect |
| add physical parameters (t, A) to action space (ablation) | Dehaze-RL · real-world hazy images | visual quality | Improved |
| add enhancement actions (S, C) (ablation) | Dehaze-RL · real-world hazy images | visual quality | Improved |
| add gating policy map g (ablation) | Dehaze-RL · real-world hazy images | visual quality | Improved |
Closest papers by meaning
- Exploiting Diffusion Prior for Real-World Image Dehazing with Unpaired Training (AAAI 2025 · similarity 0.88)
- Learning Hazing to Dehazing: Towards Realistic Haze Generation for Real-World Image Dehazing (CVPR 2025 · similarity 0.88)
- Dehaze-RetinexGAN: Real-World Image Dehazing via Retinex-based Generative Adversarial Network (AAAI 2025 · similarity 0.87)
- PromptHaze: Prompting Real-world Dehazing via Depth Anything Model (AAAI 2025 · similarity 0.86)
- Adaptive Dynamic Dehazing via Instruction-Driven and Task-Feedback Closed-Loop Optimization for Diverse Downstream Task Adaptation (AAAI 2026 · similarity 0.86)
- MaskSense: Confronting the Visual Exploration Trap in Masked Image Generation (NeurIPS 2026 · similarity 0.85)
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.
