Research map / Image generation / Image restoration
Hybrid-LUT: Channel-Aware Hybrid Lookup Table and Filtering for Efficient Image Restoration
Zhilin Ai, Boyu Li, Sidi Yang, Wenqing Shi, Wenyong Zhou, Binxiao Huang, Chenchen Ding, Ngai Wong
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 UV mean filter with ground-truth UV (oracle) (ablation) | SRLUT Y + Mean Filter UV · CBSD68 (sigma=15) | CPSNR (dB): 31.40 -> 32.15 (+0.75) | Improved |
| replace SRLUT Y output with ground-truth Y (oracle) (ablation) | SRLUT Y + Mean Filter UV · CBSD68 (sigma=15) | CPSNR (dB): 31.40 -> 39.12 (+7.72) | Improved |
| asymmetric YUV (LUT on Y, filter on UV) instead of LUT on R only with filter on GB (ablation) | Hybrid-LUT · CBSD68 (sigma=15) | CPSNR (dB): 28.11 -> 32.47 | Improved |
| asymmetric YUV instead of RGB LUT on R only (ablation) | Hybrid-LUT · Urban100 (sigma=15) | CPSNR (dB): 25.79 -> 31.17 | Improved |
| use LUTs on U and V as well as Y (symmetric) instead of mean filter (ablation) | Hybrid-LUT · CBSD68 (sigma=15) | CPSNR (dB): 32.47 -> 32.67 (1684 KB -> 3796 KB) | Improved |
| use LUTs on U and V as well as Y (symmetric) instead of mean filter (ablation) | Hybrid-LUT · McMaster (sigma=15) | CPSNR (dB): 31.78 -> 32.92 | Improved |
| use LUTs on U and V as well as Y (symmetric) instead of mean filter (ablation) | Hybrid-LUT · SIDD (real-world) | CPSNR (dB): 36.07 -> 35.62 | Hurt |
| pixel-wise learnable fusion instead of per-tensor learnable weights (ablation) | Hybrid-LUT · CBSD68 (sigma=15) | CPSNR (dB): 32.36 -> 32.47 | Improved |
| pixel-wise learnable fusion instead of per-row learnable weights (ablation) | Hybrid-LUT · CBSD68 (sigma=15) | CPSNR (dB): 32.39 -> 32.47 | Improved |
| pixel-wise learnable fusion instead of fixed average weights (ablation) | Hybrid-LUT · McMaster (sigma=15) | CPSNR (dB): 31.58 -> 31.78 | Improved |
| per-tensor learnable weights instead of fixed average (ablation) | Hybrid-LUT fusion · CBSD68 (sigma=15) | CPSNR (dB): 32.42 -> 32.36 | Hurt |
| complementary multi-unit LUT ensemble (HDBL/HD/HDBLRC and L/HD/HDBLRC) instead of single HD unit per branch (ablation) | Hybrid-LUT · McMaster (sigma=15) | CPSNR (dB): 30.83 -> 31.78 (278 KB -> 421 KB) | Improved |
Closest papers by meaning
- TinyLUT: Tiny Look-Up Table for Efficient Image Restoration at the Edge (NeurIPS 2024 · similarity 0.88)
- DnLUT: Ultra-Efficient Color Image Denoising via Channel-Aware Lookup Tables (CVPR 2025 · similarity 0.87)
- NILUT: Conditional Neural Implicit 3D Lookup Tables for Image Enhancement (AAAI 2024 · similarity 0.85)
- Look-Up Table Compression for Efficient Image Restoration (CVPR 2024 · similarity 0.85)
- PixelART: Image-to-Layer Decomposition without Latents or Text-to-Image Pretraining (NeurIPS 2026 · similarity 0.84)
- Efficient Neural Network Encoding for 3D Color Lookup Tables (AAAI 2025 · 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.
