Research map / Architectures and efficiency / CNN design
The Neural Tangent Link Between CNN Denoisers and Non-Local Filters
Check 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 the corrupted image rather than noise as input (ablation) | Vanilla CNN trained with Adam · 9 color images, Gaussian noise σ=25 | PSNR (dB): 19.6 -> 27.4 | Helped |
| Use the corrupted image rather than noise as input (ablation) | U-Net trained with Adam · 9 color images, Gaussian noise σ=25 | PSNR (dB): 28.3 -> 28.1 | Hurt |
| Use the corrupted image rather than noise as input (ablation) | Autoencoder trained with Adam · 9 color images, Gaussian noise σ=25 | PSNR (dB): 29.2 -> 29.3 | No clear effect |
| Use the corrupted image rather than noise as input (ablation) | Vanilla CNN trained with gradient descent · 9 color images, Gaussian noise σ=25 | PSNR (dB): 15.2 -> 27.5 | Helped |
| Use the corrupted image rather than noise as input (ablation) | U-Net trained with gradient descent · 9 color images, Gaussian noise σ=25 | PSNR (dB): 16.5 -> 27.1 | Helped |
| Use the corrupted image rather than noise as input (ablation) | Autoencoder trained with gradient descent · 9 color images, Gaussian noise σ=25 | PSNR (dB): 15.0 -> 26.8 | Helped |
| Replace gradient descent with Adam (ablation) | Autoencoder with noise input · 9 color images, Gaussian noise σ=25 | PSNR (dB): 15.0 -> 29.2 | Helped |
| Replace gradient descent with Adam (ablation) | Autoencoder with image input · 9 color images, Gaussian noise σ=25 | PSNR (dB): 26.8 -> 29.3 | Helped |
| Use the corrupted image rather than noise as input (ablation) | Autoencoder trained with Adam · 9 color images, Gaussian noise σ=5 | PSNR (dB): 33.5 -> 35.3 | Helped |
| Use the corrupted image rather than noise as input (ablation) | Autoencoder trained with Adam · 9 color images, Gaussian noise σ=100 | PSNR (dB): 24.4 -> 24.1 | Hurt |
| Approximate the analytic vanilla CNN filter with Nyström rather than train the CNN with gradient descent (ablation) | Vanilla CNN with image input · 9 color images, Gaussian noise σ=25 | PSNR (dB): 27.5 -> 28.3 | Helped |
| Train an autoencoder with Adam instead of using the analytic vanilla CNN filter | Image-input denoising · 9 color images, Gaussian noise σ=25 | PSNR (dB): 28.3 -> 29.3 | Helped |
Closest papers by meaning
- Universal Denoising Networks : A Novel CNN Architecture for Image Denoising (CVPR 2018 · similarity 0.90)
- Non-Local Color Image Denoising With Convolutional Neural Networks (CVPR 2017 · similarity 0.89)
- Non-Local Recurrent Network for Image Restoration (NeurIPS 2018 · similarity 0.89)
- Learning Fully Convolutional Networks for Iterative Non-Blind Deconvolution (CVPR 2017 · similarity 0.88)
- Learning Deep CNN Denoiser Prior for Image Restoration (CVPR 2017 · similarity 0.87)
- Image Restoration by Denoising Diffusion Models with Iteratively Preconditioned Guidance (CVPR 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.