Research map / Image generation / Image restoration
Denoising and Regularization via Exploiting the Structural Bias of Convolutional Generators
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 the over-parameterized deep decoder with early stopping instead of the k=128 decoder trained without early stopping (ablation) | Deep decoder · ImageNet validation, 100 images with Gaussian noise | PSNR: 27.84dB -> 28.08dB | Helped |
| Use the over-parameterized deep decoder with early stopping instead of BM3D | Deep decoder · ImageNet validation, 100 images with Gaussian noise | PSNR: 25.52 dB -> 28.08dB | Helped |
| Use fixed interpolation convolutions rather than trainable convolutional filters (ablation) | Convolutional generator fitting · Phantom MRI versus noise | Iterations to obtain a good fit | Helped |
| Increase over-parameterization of a generator with trainable convolutional filters (ablation) | Parameterized-convolution and deconvolutional generators · Phantom MRI versus noise | Difference in fitting speed | Hurt |
| Fit a natural image rather than noise with a fixed-filter convolutional generator | Convolutional generator with fixed filters · Image fitting, including ImageNet images | Fitting speed | Helped |
Closest papers by meaning
- Robust And Interpretable Blind Image Denoising Via Bias-Free Convolutional Neural Networks (ICLR 2020 · similarity 0.92)
- Universal Denoising Networks : A Novel CNN Architecture for Image Denoising (CVPR 2018 · similarity 0.90)
- Deep Image Prior (CVPR 2018 · similarity 0.89)
- Generalization in diffusion models arises from geometry-adaptive harmonic representations (ICLR 2024 · similarity 0.88)
- Denoising as Adaptation: Noise-Space Domain Adaptation for Image Restoration (ICLR 2025 · similarity 0.87)
- Factorized Diffusion Architectures for Unsupervised Image Generation and Segmentation (NeurIPS 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.