Research map / Image generation / Image restoration
On Measuring and Controlling the Spectral Bias of the Deep Image Prior
Zenglin Shi, Pascal Mettes, Subhransu Maji, Cees G. M. Snoek
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 controlled spectral bias instead of DIP with oracle stopping | Decoder with Lipschitz normalization and Gaussian upsampling · CBSD68 denoising sigma=15 | PSNR: 30.58 -> 30.80 | Improved |
| use controlled spectral bias instead of DIP with oracle stopping | Decoder with Lipschitz normalization and Gaussian upsampling · CBSD68 denoising sigma=50 | PSNR: 24.59 -> 24.83 | Improved |
| use controlled spectral bias instead of DIP with oracle stopping | Decoder with Lipschitz normalization and Gaussian upsampling · LIVE1 JPEG deblocking quality=10 | PSNR: 27.52 -> 27.70 | Improved |
| use controlled spectral bias instead of DIP with oracle stopping | Decoder with Lipschitz normalization and Gaussian upsampling · CBSD68 region inpainting ratio=0.5 | PSNR: 17.26 -> 17.36 | Improved |
| use controlled spectral bias instead of DIP with oracle stopping | Decoder with Lipschitz normalization and Gaussian upsampling · Set14 4x super-resolution | PSNR: 27.00 -> 27.50 | Improved |
| use controlled spectral bias instead of DIP with oracle stopping | Decoder with Lipschitz normalization and Gaussian upsampling · Set14 8x super-resolution | PSNR: 24.15 -> 24.17 | No clear effect |
| use controlled spectral bias instead of DIP with oracle stopping | Decoder with Lipschitz normalization and Gaussian upsampling · Set5 4x super-resolution | PSNR: 29.89 -> 30.45 | Improved |
| compare unsupervised method with supervised networks | Decoder with Lipschitz normalization and Gaussian upsampling · CBSD68 denoising sigma=25 | PSNR: 28.15 vs 31.21 (FFDNet) | Hurt |
| raise noise level to sigma=100 | Decoder with Lipschitz normalization and Gaussian upsampling · image denoising F16 | qualitative: sigma=25 works, sigma=100 fails | Hurt |
| decrease lambda in controlled prior | Decoder with Lipschitz normalization and Gaussian upsampling · image detail enhancement | smoothness: smaller lambda gives smoother output | Improved |
Closest papers by meaning
- Towards Architecture-Agnostic Untrained Networks Priors for Image Reconstruction with Frequency Regularization (ECCV 2024 · similarity 0.90)
- The Devil is in the Upsampling: Architectural Decisions Made Simpler for Denoising with Deep Image Prior (ICCV 2023 · similarity 0.89)
- A Principled Self-Referenced Early Stopping Approach for Deep Image Prior (NeurIPS 2026 · similarity 0.86)
- A Restoration Network as an Implicit Prior (ICLR 2024 · similarity 0.86)
- Untrained Neural Nets for Snapshot Compressive Imaging: Theory and Algorithms (NeurIPS 2024 · similarity 0.86)
- Image Reconstruction Via Autoencoding Sequential Deep Image Prior (NeurIPS 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.
