Research map / Image generation / Image restoration
Efficient Diffusion Model for Image Restoration by Residual Shifting
Zongsheng Yue, Jianyi Wang, Chen Change Loy
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 perceptual (LPIPS) loss (ablation) | ResShiftL · ImageNet-Test SR | LPIPS: 0.3242 -> 0.2076 | Improved |
| add perceptual (LPIPS) loss (ablation) | ResShiftL · ImageNet-Test SR | PSNR: 25.64 -> 25.02 | Hurt |
| add perceptual (LPIPS) loss (ablation) | ResShiftL · ImageNet-Test SR | MUSIQ: 41.8308 -> 51.9656 | Improved |
| replace L2 fidelity loss with L1 (ablation) | ResShiftL · ImageNet-Test SR | PSNR: 25.02 -> 24.63 | Hurt |
| replace L2 fidelity loss with L1 (ablation) | ResShiftL · ImageNet-Test SR | CLIPIQA: 0.5976 -> 0.6261 | Improved |
| replace self-attention with Swin Transformer blocks (ablation) | ResShiftL · RealSet80 | CLIPIQA: 0.5883 -> 0.6418 | Improved |
| replace self-attention with Swin Transformer blocks (ablation) | ResShiftL · ImageNet-Test | LPIPS: 0.2137 -> 0.2076 | No clear effect |
| increase diffusion steps T from 4 to 50 (ablation) | ResShift · ImageNet-Test SR | LPIPS: 0.3242 -> 0.2212 | Improved |
| increase diffusion steps T from 4 to 50 (ablation) | ResShift · ImageNet-Test SR | PSNR: 25.64 -> 24.22 | Hurt |
| increase shifting hyper-parameter p from 0.3 to 3.0 (ablation) | ResShift · ImageNet-Test SR | LPIPS: 0.2312 -> 0.3432 | Hurt |
| raise noise hyper-parameter kappa from 2.0 to 16.0 (ablation) | ResShift · ImageNet-Test SR | LPIPS: 0.2312 -> 0.2772 | Hurt |
| use DPM sampler for LDM-4 instead of DDIM (ablation) | LDM-4 · ImageNet-Test SR | LPIPS: 0.3452 -> 0.2773 | Improved |
Closest papers by meaning
- Blind Image Restoration via Fast Diffusion Inversion (NeurIPS 2024 · similarity 0.91)
- Improving Diffusion-Based Image Restoration with Error Contraction and Error Correction (AAAI 2024 · similarity 0.91)
- Residual Denoising Diffusion Models (CVPR 2024 · similarity 0.90)
- Resfusion: Denoising Diffusion Probabilistic Models for Image Restoration Based on Prior Residual Noise (NeurIPS 2024 · similarity 0.90)
- A Modular Conditional Diffusion Framework for Image Reconstruction (NeurIPS 2024 · similarity 0.90)
- Deep Equilibrium Diffusion Restoration with Parallel Sampling (CVPR 2024 · similarity 0.89)
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.
