Research map / Image generation / Image restoration
SepLUT: Separable Image-Adaptive Lookup Tables for Real-Time Image Enhancement
Canqian Yang, Meiguang Jin, Yi Xu, Rui Zhang, Ying Chen, Huaida Liu
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 |
|---|---|---|---|
| replace histogram equalization with learned 3x1D LUT as component-independent transform (ablation) | SepLUT · FiveK retouching 480p (m=6) | PSNR: 21.75 -> 25.42 | Improved |
| replace histogram equalization with learned 3x1D LUT (ablation) | SepLUT · FiveK retouching 480p (m=8) | PSNR: 21.76 -> 25.47 | Improved |
| use three per-channel 1D LUTs instead of one shared 1D LUT (ablation) | SepLUT · FiveK retouching 480p (m=6) | PSNR: 25.32 -> 25.42 | Improved |
| use three per-channel 1D LUTs instead of one shared 1D LUT (ablation) | SepLUT · FiveK retouching 480p (m=8) | PSNR: 25.41 -> 25.47 | Improved |
| add 1D LUT before 3D LUT | 3D-LUT · FiveK retouching 480p | parameters: 593.5K -> 47.2K | Improved |
| add 1D LUT before 3D LUT | 3D-LUT · FiveK retouching 480p | PSNR: 25.29 -> 25.42 | Improved |
| add 1D LUT before 3D LUT | 3D-LUT · PPR10K retouching GT a | PSNR: 25.64 -> 26.19 | Improved |
| quantize LUT generators to 8-bit (ablation) | SepLUT Ours-S · FiveK retouching 480p | PSNR: 25.42 -> 25.35 | No clear effect |
| quantize LUT generators to 8-bit (ablation) | SepLUT Ours-L · FiveK retouching 480p | parameters: 119.8K -> 76.3K | Improved |
| increase 1D LUT size beyond 3D LUT size (ablation) | SepLUT · FiveK retouching 480p | PSNR | No clear effect |
| decrease 3D LUT size from 33 to 9 (no 1D LUT) (ablation) | 3D LUT only · FiveK retouching 480p | parameters: 385K -> 69K | Improved |
| apply predicted 1D LUT (ablation) | SepLUT · FiveK histogram chi-square distance | chi-square distance: 1.6895 to 0.7520 | Improved |
Closest papers by meaning
- NILUT: Conditional Neural Implicit 3D Lookup Tables for Image Enhancement (AAAI 2024 · similarity 0.91)
- Real-Time Exposure Correction via Collaborative Transformations and Adaptive Sampling (CVPR 2024 · similarity 0.89)
- Efficient Neural Network Encoding for 3D Color Lookup Tables (AAAI 2025 · similarity 0.88)
- GLUT: 3D Gaussian Lookup Table for Continuous Color Transformation (NeurIPS 2026 · similarity 0.87)
- Pan-LUT: Efficient Pan-sharpening via Learnable Look-Up Tables (NeurIPS 2025 · similarity 0.84)
- TinyLUT: Tiny Look-Up Table for Efficient Image Restoration at the Edge (NeurIPS 2024 · similarity 0.84)
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.
