Research map / Image generation / Image restoration
Learning Distortion Invariant Representation for Image Restoration From a Causality Perspective
Xin Li, Bingchen Li, Xin Jin, Cuiling Lan, Zhibo Chen
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 |
|---|---|---|---|
| Apply DIL instead of ERM | RRDB image denoising · CBSD68, noise level 30 (unseen) | PSNR: 24.90 -> 30.29 | Improved |
| Use parallel sampling instead of serial sampling (ablation) | DIL image denoising with first-order optimization · CBSD68, noise level 30 (unseen) | PSNR: 30.29 -> 29.92 | Hurt |
| Use second-order instead of first-order optimization with serial sampling (ablation) | DIL image denoising · CBSD68, noise level 30 (unseen) | PSNR: 30.29 -> 27.48 | Hurt |
| Use parallel sampling and second-order optimization (ablation) | DIL image denoising · CBSD68, noise level 30 (unseen) | PSNR: 30.29 -> 29.14 | Hurt |
| Apply DIL instead of ERM | RRDB image denoising · Kodak24, noise level 50 (unseen) | PSNR: 19.02 -> 27.76 | Improved |
| Use DIL with serial sampling and second-order optimization (ablation) | RRDB image denoising · Manga109, noise level 30 (unseen) | PSNR: 26.62 -> 26.02 | Hurt |
| Apply DIL instead of ERM | RRDB image deblurring · Set5, blur level 5.0 (unseen) | PSNR: 22.00 -> 23.46 | Improved |
| Apply DIL instead of ERM | RRDB hybrid distortion removal · Urban100, mild distortion (unseen) | PSNR: 23.97 -> 25.00 | Improved |
| Apply DIL instead of ERM | RRDB real image super-resolution · RealSR V3 (unseen) | PSNR: 27.65 -> 27.94 | Improved |
| Use parallel sampling and second-order optimization instead of serial sampling and first-order optimization (ablation) | DIL real image super-resolution · RealSR V3 (unseen) | PSNR: 27.94 -> 28.12 | Improved |
| Apply DIL to the SwinIR backbone instead of ERM | SwinIR image denoising · CBSD68, noise level 30 (unseen) | PSNR: 24.22 -> 29.08 | Improved |
| Apply DIL with D2 rather than ERM with D2 augmentation | RRDB real image super-resolution · RealSR V3 (unseen) | PSNR: 27.39 -> 27.65 | Improved |
Closest papers by meaning
- Denoising Prior Driven Deep Neural Network for Image Restoration (TPAMI 2018 · similarity 0.88)
- Neural Sparse Representation for Image Restoration (NeurIPS 2020 · similarity 0.88)
- On-Demand Learning for Deep Image Restoration (ICCV 2017 · similarity 0.88)
- Learning Invariant Representation for Unsupervised Image Restoration (CVPR 2020 · similarity 0.87)
- Degradation-Aware Feature Perturbation for All-in-One Image Restoration (CVPR 2025 · similarity 0.86)
- A Restoration Network as an Implicit Prior (ICLR 2024 · 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.
