Research map / Probabilistic and scientific modeling / Differential equations and PDEs
HazeFlow: Revisit Haze Physical Model as ODE and Non-Homogeneous Haze Generation for Real-World Dehazing
Junseong Shin, Seungwoo Chung, Yunjeong Yang, Tae Hyun Kim
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 Reflow stage after pretraining (ablation) | HazeFlow · RTTS | BRISQUE: 18.96 -> 8.14 | Improved |
| add Reflow stage after pretraining (ablation) | HazeFlow · RTTS | FADE: 0.561 -> 0.598 | Hurt |
| add Distillation stage after Reflow (ablation) | HazeFlow · RTTS | BRISQUE: 8.14 -> 5.01 | Improved |
| train with MCBM non-homogeneous haze synthesis instead of prior haze synthesis (ablation) | HazeFlow · RTTS | FADE: 0.950 -> 0.817 | Improved |
| add transmission refinement (T-Refinement) (ablation) | HazeFlow · RTTS | FADE: 0.817 -> 0.583 | Improved |
| add T-Refinement (ablation) | HazeFlow · RTTS | BRISQUE: 12.67 -> 5.01 | Improved |
| use MCBM haze generation to train other dehazing backbones versus RIDCP augmentation | DehazeFormer, NAFNet, FocalNet · RTTS | FADE: DehazeFormer 1.44 -> 1.05; NAFNet 1.17 -> 1.03; FocalNet 1.36 -> 0.80 | Improved |
| use MCBM haze generation instead of RIDCP augmentation | NAFNet and FocalNet · RTTS | BRISQUE: NAFNet 21.01 -> 23.68; FocalNet 28.07 -> 29.90 | Mixed |
| use ASM-derived ODE instead of naive rectified flow linear interpolation (ablation) | HazeFlow · RTTS | FADE: 1.059 -> 0.583 | Improved |
| use one-step flow inference instead of iterative diffusion sampling | HazeFlow vs diffusion dehazers · RTTS | run time (s): 38.09 (DACLIP, 100 steps) -> 0.21 (1 step) | Improved |
| increase number of estimation steps at inference | HazeFlow · RTTS | FADE/BRISQUE vs PAQ2PIQ | Mixed |
| replace DCP with Non-local or DCPDN for initial transmission map (ablation) | HazeFlow · RTTS | FADE: 0.583 -> 0.734 (Non-local); 0.583 -> 0.713 (DCPDN) | Hurt |
Closest papers by meaning
- Learning Hazing to Dehazing: Towards Realistic Haze Generation for Real-World Image Dehazing (CVPR 2025 · similarity 0.91)
- PRISM: Rethinking Atmospheric Scattering Reconstruction as a Unified Understanding and Restoration Model for Real-world Dehazing (NeurIPS 2026 · similarity 0.91)
- Real-world Image Dehazing with Coherence-based Pseudo Labeling and Cooperative Unfolding Network (NeurIPS 2024 · similarity 0.88)
- Tokenize Image Patches: Global Context Fusion for Effective Haze Removal in Large Images (CVPR 2025 · similarity 0.87)
- PEIE: Physics Embedded Illumination Estimation for Adaptive Dehazing (AAAI 2025 · similarity 0.87)
- PromptHaze: Prompting Real-world Dehazing via Depth Anything Model (AAAI 2025 · similarity 0.87)
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.
