Research map / Image generation / Image restoration
Learning Pixel-adaptive Multi-layer Perceptrons for Real-time Image Enhancement
Junyu Lou, Xiaorui Zhao, Kexuan Shi, Shuhang Gu
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 affine transform with pixel-adaptive MLP (ablation) | BPAM · FiveK tone mapping (480p) | PSNR: 25.53 -> 25.70 | Improved |
| replace affine transform with pixel-adaptive MLP (with grid decomposition) (ablation) | BPAM · FiveK tone mapping (480p) | PSNR: 25.63 -> 25.83 | Improved |
| add grid decomposition to the MLP (ablation) | BPAM · FiveK tone mapping (480p) | PSNR: 25.70 -> 25.83 | Improved |
| add grid decomposition to the affine transform (ablation) | BPAM · FiveK tone mapping (480p) | PSNR: 25.53 -> 25.63 | Improved |
| use multi-channel guidance map (scheme 2) instead of per-channel input guidance (scheme 1) (ablation) | BPAM · FiveK tone mapping (480p) | PSNR: 25.78 -> 25.83 | Improved |
| widen MLP hidden layer from 8 to 16 (ablation) | BPAM · FiveK tone mapping (480p) | PSNR: 25.83 -> 25.81 | No clear effect |
| add a second hidden layer to the MLP (3-8-8-3) (ablation) | BPAM · FiveK tone mapping (480p) | PSNR: 25.83 -> 25.87 | No clear effect |
| use BPAM instead of LutBGrid | BPAM · FiveK tone mapping (full resolution) | PSNR: 24.57 -> 25.12 | Improved |
| use BPAM instead of LutBGrid | BPAM · FiveK photo retouching (480p) | PSNR: 25.66 -> 25.73 | Improved |
| use BPAM instead of CoTF | BPAM · LCDP exposure correction | PSNR: 23.89 -> 24.22 | Improved |
| use BPAM instead of 3DLUT in speed | BPAM · Inference time 3840x2160 | time (ms): 1.16 -> 27.8 | Hurt |
| use BPAM instead of PLSA | BPAM · PPR10K-b photo retouching | PSNR: 25.51 -> 25.57 | Improved |
Closest papers by meaning
- Image-adaptive 3D Lookup Tables for Real-time Image Enhancement with Bilateral Grids (ECCV 2024 · similarity 0.85)
- MAXIM: Multi-Axis MLP for Image Processing (CVPR 2022 · similarity 0.85)
- Real-Time Image Enhancer via Learnable Spatial-Aware 3D Lookup Tables (ICCV 2021 · similarity 0.84)
- PixelART: Image-to-Layer Decomposition without Latents or Text-to-Image Pretraining (NeurIPS 2026 · similarity 0.84)
- Coordinate-Aware Modulation for Neural Fields (ICLR 2024 · similarity 0.83)
- L2P: Unlocking Latent Potential for Pixel Generation (NeurIPS 2026 · 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.
