Research map / Image generation / Image restoration
xUnit: Learning a Spatial Activation Function for Efficient Image Restoration
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 |
|---|---|---|---|
| Replace ReLUs with xUnits and reduce the denoising network from 17 to 9 convolutional layers | DnCNN-S · BSD68, σ=25 | PSNR [dB]: 29.22 -> 29.21 | No clear effect |
| Replace ReLUs with xUnits and reduce the denoising network from 17 to 9 convolutional layers | DnCNN-S · BSD68, σ=50 | PSNR [dB]: 26.23 -> 26.26 | Helped |
| Use xUnit activations in the deraining network | DnCNN · Rain12 | PSNR [dB]: 30.90 -> 31.17 | Helped |
| Replace ReLUs with xUnits and reduce middle-layer filter size (ablation) | SRCNN · BSD100, 3× | PSNR [dB]: 28.41 -> 28.53 | Helped |
| Replace ReLUs with xUnits and reduce middle-layer channels (ablation) | SRCNN · BSD100, 3× | PSNR [dB]: 28.41 -> 28.54 | Helped |
| Replace residual-block PReLUs with xUnits and reduce residual blocks from 16 to 10 | SRResNet · BSD100, 4× | PSNR [dB]: 27.58 -> 27.61 | Helped |
| Use xUnit activations and reduce middle-layer filter size (ablation) | SRCNN · BSD100, 4× | PSNR [dB]: 26.90 -> 27.06 | Helped |
| Use xUnit activations and reduce middle-layer channels (ablation) | SRCNN · BSD100, 4× | PSNR [dB]: 26.90 -> 27.04 | Helped |
| Use spatial depth-wise filters in the xUnit rather than 1×1 filters (ablation) | xNet · BSD68, σ=25 | PSNR | Helped |
| Invest parameters in xUnit activations rather than allocating 0% of parameters to activations (ablation) | xNet · BSD68, σ=25 | PSNR | Helped |
| Use the Gaussian gating function to keep activation weights in [0,1] (ablation) | xUnit · Denoising, deeper networks | training stability | Helped |
| Remove the Gaussian gating function from the BN+RL+CD design (ablation) | xUnit · Denoising, deeper networks | training stability | Hurt |
Closest papers by meaning
- Omni-Kernel Network for Image Restoration (AAAI 2024 · similarity 0.89)
- Adapt or Perish: Adaptive Sparse Transformer with Attentive Feature Refinement for Image Restoration (CVPR 2024 · similarity 0.87)
- Efficient Visual State Space Model for Image Deblurring (CVPR 2025 · similarity 0.87)
- TinyLUT: Tiny Look-Up Table for Efficient Image Restoration at the Edge (NeurIPS 2024 · similarity 0.87)
- Improving the Learning Capability of Small-size Image Restoration Network by Deep Fourier Shifting (NeurIPS 2024 · similarity 0.87)
- A Restoration Network as an Implicit Prior (ICLR 2024 · similarity 0.86)
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.