Research map / Image generation / Image restoration
Neural Color Operators for Sequential Image Retouching
Yili Wang, Xin Li, Kun Xu, Dongliang He, Qi Zhang, Fu Li, Errui Ding
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 |
|---|---|---|---|
| initialize neurOps with standard color operators instead of random (ablation) | NeurOp · MIT-Adobe-5K-Dark | PSNR: 22.61 -> 24.32 | Improved |
| allow finetuning of standard-operator-initialized neurOps (ablation) | NeurOp · MIT-Adobe-5K-Dark | PSNR: 23.78 -> 24.32 | Improved |
| use K=3 neurOps instead of 1 (ablation) | NeurOp · MIT-Adobe-5K-Dark | PSNR: 21.97 -> 24.32 | Improved |
| use K=3 neurOps instead of 2 (ablation) | NeurOp · MIT-Adobe-5K-Dark | PSNR: 23.5 -> 24.32 | Improved |
| use K=3 neurOps instead of 4 (ablation) | NeurOp · MIT-Adobe-5K-Dark | PSNR: 24.17 -> 24.32 | Improved |
| order black clipping, exposure, vibrance for initialization instead of vibrance, black clipping, exposure (ablation) | NeurOp · MIT-Adobe-5K-Dark | PSNR: 23.93 -> 24.32 | Improved |
| add max and std pooling to average pooling (ablation) | Strength predictor · MIT-Adobe-5K-Dark | PSNR: 23.96 -> 24.32 | Improved |
| feed intermediate images instead of the original input to strength predictors (ablation) | NeurOp · MIT-Adobe-5K-Dark | PSNR: 23.91 -> 24.32 | Improved |
| do not share conv parameters among strength predictors (ablation) | NeurOp · MIT-Adobe-5K-Dark | PSNR: 24.32 -> 24.35 | No clear effect |
| add TV loss and color loss to reconstruction loss (ablation) | NeurOp · MIT-Adobe-5K-Dark | PSNR: 24.22 -> 24.32 | Improved |
| use NeurOp instead of CSRNet | CSRNet · PPR10K-a | PSNR: 24.24 -> 26.32 | Improved |
| add human region priority (HRP) training | NeurOp · PPR10K-b | PSNR: 25.45 -> 25.82 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DeepLPF — DeepLPF: Deep Local Parametric Filters for Image Enhancement (CVPR 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Distort-and-Recover: Color Enhancement Using Deep Reinforcement Learning (CVPR 2018 · similarity 0.87)
- Conditional Sequential Modulation for Efficient Global Image Retouching (ECCV 2020 · similarity 0.86)
- Neural Preset for Color Style Transfer (CVPR 2023 · similarity 0.86)
- ColorFormer: Image Colorization via Color Memory Assisted Hybrid-Attention Transformer (ECCV 2022 · similarity 0.86)
- Close Imitation of Expert Retouching for Black-and-White Photography (CVPR 2024 · similarity 0.85)
- Automatic Controllable Colorization via Imagination (CVPR 2024 · similarity 0.83)
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.
