Research map / Image generation / Image restoration
Image-adaptive 3D Lookup Tables for Real-time Image Enhancement with Bilateral Grids
Wontae Kim, Nam Ik Cho
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 |
|---|---|---|---|
| Use bilateral grids to add spatial features to a 3D LUT enhancer | adaptive 3D LUT · FiveK photo retouch (480p) | PSNR: 25.49 -> 25.66 (vs AdaInt) | Improved |
| Replace learned guidance map generator with raw input image channels (ablation) | bilateral grid slicing · FiveK photo retouch (480p) | PSNR: 23.42 -> 25.66 (1x1 generator vs input) | Improved |
| Use input image as guidance map instead of generator (inference time) (ablation) | bilateral grid slicing · FiveK inference time | inference time (ms): generators cost 21.6ms and 46.7ms | Improved |
| Add perceptual loss Lp (ablation) | full model · FiveK photo retouch (480p) | PSNR: 25.21 -> 25.49 | Improved |
| Add color difference loss Lc on top of Lb and Lp (ablation) | full model · FiveK photo retouch (480p) | PSNR: 25.49 -> 25.66 | Improved |
| Use trilinear interpolation for slicing and tetrahedral for 3D LUT instead of trilinear for both (ablation) | full model · FiveK photo retouch (480p) | PSNR: 25.57 -> 25.66 | Improved |
| Use tetrahedral instead of trilinear interpolation for bilateral grid slicing (ablation) | full model · FiveK photo retouch (480p) | PSNR: 25.66 -> 25.50 | Hurt |
| Increase LUTGen hidden layer from Mt=8 to Mt=24 (ablation) | LUTGen · FiveK photo retouch (480p) | PSNR: +0.02dB for 239.9K more parameters | No clear effect |
| Increase GridGen hidden layer to Mb=32 (ablation) | GridGen · FiveK photo retouch (480p) | PSNR: 0.01dB up for 359.9K more parameters | No clear effect |
| Apply adaptive sampling to the 3D LUT (ablation) | full model · FiveK photo retouch | PSNR: -0.15dB | Hurt |
| Add local features to bilateral grid generation (ablation) | GridGen · FiveK photo retouch | PSNR: PSNR of 25.30 dB | No clear effect |
| Combine both bilateral grid and 3D LUT modules (vs either alone) (ablation) | full model · FiveK photo retouch (480p) | PSNR | Improved |
Closest papers by meaning
- Real-Time Exposure Correction via Collaborative Transformations and Adaptive Sampling (CVPR 2024 · similarity 0.89)
- NILUT: Conditional Neural Implicit 3D Lookup Tables for Image Enhancement (AAAI 2024 · similarity 0.88)
- TinyLUT: Tiny Look-Up Table for Efficient Image Restoration at the Edge (NeurIPS 2024 · similarity 0.84)
- Efficient Neural Network Encoding for 3D Color Lookup Tables (AAAI 2025 · similarity 0.84)
- Pan-LUT: Efficient Pan-sharpening via Learnable Look-Up Tables (NeurIPS 2025 · similarity 0.83)
- Look-Up Table Compression for Efficient Image Restoration (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.
