Research map / Architectures and efficiency / CNN design
Basic Binary Convolution Unit for Binarized Image Restoration Network
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 BBCU-L instead of ReActNet | SRResNet body, ×2 super-resolution · Set5, Urban100, and Manga109 | PSNR: +0.32dB; +0.27dB; +0.58dB | Helped |
| Use BBCU-L instead of ReActNet | SRResNet body, ×4 super-resolution · Set5, Urban100, and Manga109 | PSNR: 0.25dB, 0.27dB, and 0.44dB | Helped |
| Use BBCU instead of ReActNet | DnCNN body for denoising · CBSD68, Kodak24, and Urban100, σ=15 | PSNR: 0.82dB, 0.88dB, and 1dB | Helped |
| Use BBCU instead of ReActNet | DnCNN-3 body for JPEG artifact reduction · Live1 and Classic5, q=40 | PSNR: 0.38dB and 0.34dB | Helped |
| Remove BatchNorm (ablation) | BBCU-V1 · 4× SR, Urban100 and Manga109 | PSNR: 25.35 -> 25.21; 29.25 -> 29.01 | Hurt |
| Add residual alignment to the no-BatchNorm BBCU (ablation) | BBCU-V2 · 4× SR, Set5, Urban100, and Manga109 | PSNR: 0.17dB, 0.16dB, and 0.29dB | Helped |
| Move the activation function into the residual connection (ablation) | BBCU-V3 · 4× SR | PSNR: 31.71 -> 31.79; 25.51 -> 25.62; 29.54 -> 29.69 | Helped |
| Add a residual connection to each binary convolution (ablation) | Binarized SRResNet · 4× Urban100 | PSNR: 25.43; 25.44; 25.48; 25.46; 25.45; 25.44 -> 25.62 | Helped |
| Vary the residual breakpoint position (ablation) | Binarized SRResNet · 4× Urban100 | PSNR: 25.43; 25.44; 25.48; 25.46; 25.45; 25.44 | No clear effect |
| Tune the residual-alignment amplification factor to channel count (ablation) | Binarized SRResNet · 4× Urban100 | PSNR | Mixed |
| Binarize the upsampling part (ablation) | SRResNet · Set14 ×4 | PSNR: 28.60 -> 28.59 | Hurt |
| Binarize the head part (ablation) | SRResNet · Set14 ×4 | PSNR: 28.60 -> 28.58 | Hurt |
Closest papers by meaning
- BD-Net: Has Depth-Wise Convolution Ever Been Applied in Binary Neural Networks? (AAAI 2026 · similarity 0.90)
- Fast and Accurate Binary Neural Networks Based on Depth-Width Reshaping (AAAI 2023 · similarity 0.87)
- Sparsity-Inducing Binarized Neural Networks (AAAI 2020 · similarity 0.87)
- IRNeXt: Rethinking Convolutional Network Design for Image Restoration (ICML 2023 · similarity 0.87)
- Training binary neural networks with real-to-binary convolutions (ICLR 2020 · similarity 0.87)
- Binarized Low-light Raw Video Enhancement (CVPR 2024 · similarity 0.85)
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.