Research map / Image generation / GANs
Physics-Based Generative Adversarial Models for Image Restoration and Beyond
Jinshan Pan, Jiangxin Dong, Yang Liu, Jiawei Zhang, Jimmy Ren, Jinhui Tang, Yu‐Wing Tai, Ming–Hsuan Yang
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 the physics-model constraint (second discriminator and re-degradation losses) to the GAN (ablation) | BaseGAN · Image dehazing ablation | PSNR: 29.90 -> 32.05 | Improved |
| add the physics-model constraint to the GAN (ablation) | BaseGAN · Image dehazing ablation | SSIM: 0.9541 -> 0.9722 | Improved |
| add the second ablated loss term (ablation) | Ours · Image dehazing ablation | PSNR: 21.24 -> 32.05 | Improved |
| add the first ablated loss term (ablation) | Ours · Image dehazing ablation | PSNR: 31.93 -> 32.05 | Improved |
| add the third ablated loss term (ablation) | Ours · Image dehazing ablation | PSNR: 31.71 -> 32.05 | Improved |
| add both of the paired loss terms (ablation) | Ours · Image dehazing ablation | PSNR: 30.53 -> 32.05 | Improved |
| use physics-model GAN instead of PCycleGAN with same losses (ablation) | PCycleGAN · Image dehazing ablation | PSNR: 27.46 -> 32.05 | Improved |
| use physics model rather than cycle consistency | pix2pix · Dehazing synthetic | SSIM: 0.8225 -> 0.8657 | Improved |
| use physics model rather than feed-forward CNN | CNN · Text image deblurring | PSNR: 26.53 -> 28.80 | Improved |
| raise lambda from 10 to 50 (ablation) | Ours · Face deblurring | PSNR: 22.92 -> 24.57 | Improved |
| raise lambda from 50 to 100 (ablation) | Ours · Face deblurring | PSNR: 24.57 -> 24.55 | No clear effect |
| increase ResBlocks from 3 to 35 (ablation) | Ours · Face deblurring | PSNR: 20.45 -> 21.22 | No clear effect |
Closest papers by meaning
- Learning to Super-Resolve Blurry Face and Text Images (ICCV 2017 · similarity 0.88)
- Robustness and Generalization via Generative Adversarial Training (ICCV 2021 · similarity 0.88)
- Re-Boosting Self-Collaboration Parallel Prompt GAN for Unsupervised Image Restoration (TPAMI 2025 · similarity 0.88)
- Invertible generative models for inverse problems: mitigating representation error and dataset bias (ICML 2020 · similarity 0.88)
- Single Image Dehazing via Conditional Generative Adversarial Network (CVPR 2018 · similarity 0.88)
- Diffusion-based Adversarial Purification from the Perspective of the Frequency Domain (ICML 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.
