Research map / Image generation / Image restoration
Path-Restore: Learning Network Path Selection for Image Restoration
Ke Yu, Xintao Wang, Chao Dong, Xiaoou Tang, 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 pathfinder for dynamic path selection (ablation) | Path-Restore-Mask · DND | time (s): 0.199 -> 0.143 | Improved |
| add pathfinder for dynamic path selection (ablation) | Path-Restore-Mask · SIDD | PSNR (dB): 38.23 -> 38.50 | Improved |
| add RIDNet's larger training data (ablation) | Path-Restore-Mask · DND | PSNR (dB): 39.18 -> 39.37 | Improved |
| add RIDNet's larger training data (ablation) | Path-Restore-Mask · SIDD | PSNR (dB): 38.42 -> 38.50 | Improved |
| use 4 paths instead of 2 paths per dynamic block (ablation) | Path-Restore · DIV2K-T50 (2K image) | FLOPs (G): 1.48 -> 1.38 | Improved |
| use mask-based pathfinder over whole image (Path-Restore-Mask) (ablation) | Path-Restore · DND | PSNR (dB): 39.00 -> 39.18 | Improved |
| use mask-based pathfinder over whole image (Path-Restore-Mask) (ablation) | Path-Restore · SIDD | PSNR (dB): 38.21 -> 38.42 | Improved |
| add difficulty-regulated reward instead of non-regulated reward (d=1) (ablation) | Path-Restore · DIV2K-T50 linear noise | PSNR vs FLOPs trade-off: orange curve below blue curve (Fig) | Improved |
| use learned pathfinder instead of random path (ablation) | Path-Restore · DIV2K-T50 mixed distortions | PSNR (dB) | Improved |
| use Path-Restore dynamic paths instead of DnCNN at same FLOPs (ablation) | DnCNN · Gaussian denoising | PSNR (dB): 0.1 dB higher | Improved |
| use Path-Restore instead of DnCNN (ablation) | DnCNN · CBSD68 uniform sigma=10 | FLOPs (G): 5.31 -> 4.22 | Improved |
| deeper network, more data, lower penalty, larger patch (Path-Restore-Ext) (ablation) | Path-Restore · DND | PSNR (dB): 39.00 -> 39.72 | Improved |
Closest papers by meaning
- Revitalizing Convolutional Network for Image Restoration (TPAMI 2024 · similarity 0.91)
- Crafting a Toolchain for Image Restoration by Deep Reinforcement Learning (CVPR 2018 · similarity 0.89)
- Image Restoration Using Very Deep Convolutional Encoder-Decoder Networks with Symmetric Skip Connections (NeurIPS 2016 · similarity 0.89)
- Denoising Prior Driven Deep Neural Network for Image Restoration (TPAMI 2018 · similarity 0.89)
- Residual-Conditioned Optimal Transport: Towards Structure-Preserving Unpaired and Paired Image Restoration (ICML 2024 · similarity 0.88)
- Omni-Kernel Network for Image Restoration (AAAI 2024 · 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.
