Research map / Image generation / Image restoration
LogicIR: Logic Gate Networks for Image Restoration
Hongjae Lee, Myungjun Son, Jaeseong Yu, Seung-Won Jung
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 |
|---|---|---|---|
| add differentiable bit decoding to StackedCLGN (ablation) | StackedCLGN · Set12 denoising | PSNR: 17.19 -> 26.83 | Improved |
| add UNet backbone (ablation) | LogicIR-S · Set12 denoising | PSNR: 26.83 -> 27.15 | Improved |
| add index shuffling (ablation) | LogicIR-S · Set12 denoising | PSNR: 27.15 -> 27.58 | Improved |
| add MSB loss (ablation) | LogicIR-S · Set12 denoising | PSNR: 27.58 -> 27.72 | Improved |
| add STE fine-tuning (ablation) | LogicIR-S · Set12 denoising | PSNR: 27.72 -> 27.83 | Improved |
| add 4-rotation ensemble | LogicIR-S · BSD68 denoising (sigma=25) | PSNR: 27.40 -> 27.71 | Improved |
| widen network from S to L (C=2048 to 4096) | LogicIR · BSD68 denoising (sigma=25) | PSNR: 27.40 -> 27.50 | Improved |
| process RGB jointly (width 3C) instead of per-channel (ablation) | LogicIR-S · Test100 deraining | PSNR: 22.75 -> 22.79 | No clear effect |
| process RGB channels instead of Y-only (ablation) | LogicIR-S · Test100 deraining | PSNR: 22.63 -> 22.75 | Improved |
| keep first and last layers binary (fully binarized) vs full precision, BBCU-lite (ablation) | BBCU-lite · BSD68 denoising (sigma=25) | PSNR: 27.62 -> 25.23 | Hurt |
| use LogicIR-S instead of BBCU-lite for runtime | LogicIR-S · Denoising complexity | runtime: 571.6 ms -> 28.2 ms | Improved |
| use LogicIR-S instead of BBCU-lite for energy | LogicIR-S · Denoising complexity | energy: 0.69 mJ -> 0.03 mJ | Improved |
Closest papers by meaning
- A Comparative Study of Image Restoration Networks for General Backbone Network Design (ECCV 2024 · similarity 0.85)
- CLEARER: Multi-Scale Neural Architecture Search for Image Restoration (NeurIPS 2020 · similarity 0.85)
- Training Your Image Restoration Network Better with Random Weight Network as Optimization Function (NeurIPS 2023 · similarity 0.84)
- TinyLUT: Tiny Look-Up Table for Efficient Image Restoration at the Edge (NeurIPS 2024 · similarity 0.84)
- MambaIRv2: Attentive State Space Restoration (CVPR 2025 · similarity 0.83)
- DreamClear: High-Capacity Real-World Image Restoration with Privacy-Safe Dataset Curation (NeurIPS 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.
