Research map / Image generation / Image restoration
Stabilizing Deep Reconstruction Operators with Contractive Anchoring
Arghya Sinha, Trishit Mukherjee, Kunal Chaudhury
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 |
|---|---|---|---|
| add contractive anchoring to vanilla PnP-HQS | PnP-HQS + GS-DRUNet · CBSD10 motion deblurring | final PSNR (dB): 25.72 -> 30.70 | Improved |
| add contractive anchoring to PnP-HQS + DiffUNet | PnP-HQS + DiffUNet · CBSD10 motion deblurring | final PSNR (dB): 23.12 -> 29.61 | Improved |
| add contractive anchoring to PnP-ADMM + MMO | PnP-ADMM + MMO · CBSD10 motion deblurring | final PSNR (dB): 22.43 -> 29.09 | Improved |
| add contractive anchoring, 2x superresolution | PnP-HQS + GS-DRUNet · CBSD10 2x SR | final PSNR (dB): 23.84 -> 27.95 | Improved |
| add contractive anchoring, peak PSNR on deblurring | PnP-HQS + GS-DRUNet · CBSD10 motion deblurring | peak PSNR (dB): 30.56 -> 30.70 | Improved |
| stabilization vs Equivariant PnP on final PSNR | PnP · CBSD10 motion deblurring | final PSNR (dB): 26.10 -> 30.70 | Improved |
| add stabilization on PnP-ADMM + MMO, peak PSNR | PnP-ADMM + MMO · CBSD10 deblurring (kernel 1) | peak PSNR (dB): 27.93 -> 27.90 | No clear effect |
| use discrepancy-principle early stopping instead of peak iterate (ablation) | PnP + DRUNet · set3c Butterfly deblurring | PSNR (dB): 29.54 -> 25.04 | Hurt |
| use anchor with converged denoiser (>=300 epochs) (ablation) | stabilization anchor · divergent IR process | collapse prevention | Improved |
Closest papers by meaning
- i-DEQ: A stable inertial Deep Equilibrium model for image restoration (NeurIPS 2026 · similarity 0.87)
- Learning Pseudo-Contractive Denoisers for Inverse Problems (ICML 2024 · similarity 0.87)
- Image Restoration by Denoising Diffusion Models with Iteratively Preconditioned Guidance (CVPR 2024 · similarity 0.86)
- Deep Equilibrium Diffusion Restoration with Parallel Sampling (CVPR 2024 · similarity 0.85)
- Proximal Algorithm Unrolling: Flexible and Efficient Reconstruction Networks for Single-Pixel Imaging (CVPR 2025 · similarity 0.84)
- A Restoration Network as an Implicit Prior (ICLR 2024 · similarity 0.84)
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.
