Research map / Image generation / Image restoration
Reverse Convolution and Its Applications to Image Restoration
Xuhong Huang, Shiqi Liu, Kai Zhang, Ying Tai, Jian Yang, Hui Zeng, Lei Zhang
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 Gaussian+Softmax kernel init instead of Gaussian (ablation) | Converse-DnCNN · Set12 | PSNR (dB): 30.60 -> 30.70 | Improved |
| use Gaussian instead of uniform kernel init (ablation) | Converse-DnCNN · Set12 | PSNR (dB): 30.31 -> 30.60 | Improved |
| use 5x5 kernel instead of 3x3 (ablation) | Converse-DnCNN · Set12 | PSNR (dB): 30.58 -> 30.70 | Improved |
| use 7x7 kernel instead of 5x5 (ablation) | Converse-DnCNN · Set12 | PSNR (dB): 30.70 -> 30.71 | No clear effect |
| use circular padding instead of zero padding (ablation) | Converse-DnCNN · Set12 | PSNR (dB): 30.63 -> 30.70 | Improved |
| use padding size 4 instead of 0 (ablation) | Converse-DnCNN · Set12 | PSNR (dB): 30.67 -> 30.70 | No clear effect |
| initialize X0 with Y instead of zeros (ablation) | Converse-DnCNN · Set12 | PSNR (dB): 30.66 -> 30.70 | Improved |
| initialize X0 with Interp(Y,s) instead of zeros (ablation) | Converse-SRResNet · Set5 (x4 SR) | PSNR (dB): 32.22 -> 32.25 | Improved |
| replace depthwise convolution with Converse2D | Converse-DnCNN · Set12 | PSNR (dB): 30.64 -> 30.70 | Improved |
| replace depthwise transposed convolution with Converse2D | Converse-DnCNN · BSD68 | PSNR (dB): 29.29 -> 29.36 | Improved |
| replace depthwise convolution with Converse2D | SRResNet · Urban100 (x4 SR) | PSNR (dB): 26.24 -> 26.24 | No clear effect |
| replace depthwise transposed convolution with Converse2D | SRResNet · Urban100 (x4 SR) | PSNR (dB): 25.89 -> 26.24 | Improved |
Closest papers by meaning
- Revitalizing Convolutional Network for Image Restoration (TPAMI 2024 · similarity 0.89)
- Dual Residual Networks Leveraging the Potential of Paired Operations for Image Restoration (CVPR 2019 · similarity 0.89)
- Image Restoration Using Very Deep Convolutional Encoder-Decoder Networks with Symmetric Skip Connections (NeurIPS 2016 · similarity 0.89)
- Deep Decoder: Concise Image Representations from Untrained Non-convolutional Networks (ICLR 2019 · similarity 0.88)
- Universal Denoising Networks : A Novel CNN Architecture for Image Denoising (CVPR 2018 · similarity 0.88)
- A Restoration Network as an Implicit Prior (ICLR 2024 · similarity 0.87)
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.
