Research map / Label-efficient and robust learning / Self-supervised and contrastive
Unsupervised Deraining: Where Contrastive Learning Meets Self-Similarity
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use NLCL instead of CUT | Unsupervised deraining on RainCityscapes · RainCityscapes | PSNR: 25.21 -> 26.46 | Helped |
| Use NLCL instead of UDGNet | Unsupervised deraining on RainCityscapes · RainCityscapes | SSIM: 0.8749 -> 0.8666 | Hurt |
| Use NLCL instead of CycleGAN | Unsupervised deraining on RainCityscapes · RainCityscapes | NIQE: 3.68 -> 3.67 | Helped |
| Use NLCL instead of DSC | Unsupervised deraining on SPA · SPA | PSNR: 33.71 -> 33.82 | Helped |
| Use NLCL instead of CycleGAN | Unsupervised deraining on SPA · SPA | NIQE: 6.67 -> 9.55 | Hurt |
| Use non-local sampling for both positives and negatives rather than random sampling (ablation) | Layer contrastive in NLCL · RainCityscapes | PSNR: 25.83 -> 26.46 | Helped |
| Use non-local sampling for positives rather than neighbour sampling, with random negatives (ablation) | Layer contrastive in NLCL · RainCityscapes | PSNR: 26.03 -> 26.18 | Helped |
| Use non-local sampling for negatives rather than random sampling, with random positives (ablation) | Layer contrastive in NLCL · RainCityscapes | PSNR: 25.83 -> 26.16 | Helped |
| Use the discriminator rather than the image generator as layer-contrastive encoder (ablation) | Layer contrastive in NLCL · RainCityscapes | PSNR: 24.86 -> 26.46 | Helped |
| Use the discriminator rather than separate image and rain generator encoders (ablation) | Layer contrastive in NLCL · RainCityscapes | PSNR: 24.12 -> 26.46 | Helped |
| Use 8 positives and 256 negatives instead of the tested alternative sample counts (ablation) | Non-local layer sampling in NLCL · RainCityscapes | PSNR: 23.21; 26.11; 26.19; 26.30; 24.55; 26.30; 26.46; 26.42; 24.54; 25.96; 26.02; 26.11; 23.49; 25.02; 25.40; 25.37 | Mixed |
| Use reverse non-local instead of random location-negative sampling (ablation) | Location contrast in NLCL · RainCityscapes | PSNR: 26.33 -> 26.46 | Helped |
Closest papers by meaning
- Multi-Decoding Deraining Network and Quasi-Sparsity Based Training (CVPR 2021 · similarity 0.92)
- Channel Consistency Prior and Self-Reconstruction Strategy Based Unsupervised Image Deraining (CVPR 2025 · similarity 0.90)
- Harnessing Joint Rain-/Detail-aware Representations to Eliminate Intricate Rains (ICLR 2024 · similarity 0.89)
- Rethinking Multi-Scale Representations in Deep Deraining Transformer (AAAI 2024 · similarity 0.88)
- Bidirectional Multi-Scale Implicit Neural Representations for Image Deraining (CVPR 2024 · similarity 0.86)
- Towards Universal Rainy Image Restoration: Benchmark and Baseline (AAAI 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.