Research map / Label-efficient and robust learning / Self-supervised and contrastive
Frozen CLIP Priors for Robust Self-Supervised Poisson Inverse Problems
Laura C. Diaz-Delgado, Emmanuel Martinez, Henry Arguello
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 EI regularization to GR2R loss (ablation) | Frozen CLIP unrolled solver · Poisson demosaicing BSDS500 (gamma=0.01) | PSNR (dB): 27.45 -> 30.53 | Improved |
| add GR2R re-corruption loss to measurement consistency (ablation) | Frozen CLIP unrolled solver · Poisson demosaicing BSDS500 (gamma=0.01) | PSNR (dB): 21.42 -> 27.45 | Improved |
| use GR2R+EI self loss instead of EI only (ablation) | Frozen CLIP unrolled solver · Poisson demosaicing BSDS500 (gamma=0.01) | PSNR (dB): 28.41 -> 30.53 | Improved |
| use self loss instead of REI (ablation) | Frozen CLIP unrolled solver · Poisson demosaicing BSDS500 (gamma=0.01) | PSNR (dB): 30.31 -> 30.53 | Improved |
| use frozen pretrained CLIP encoder instead of DnCNN backbone (ablation) | Unrolled solver with L_self · Poisson demosaicing BSDS500 (gamma=0.01) | PSNR (dB): 29.90 -> 30.53 | Improved |
| use frozen CLIP instead of DRUNet backbone (ablation) | Unrolled solver with L_self · Poisson demosaicing BSDS500 (gamma=0.01) | PSNR (dB): 29.82 -> 30.53 | Improved |
| add pretraining to DnCNN backbone (ablation) | DnCNN + L_self · Poisson demosaicing BSDS500 (gamma=0.01) | PSNR (dB): 29.90 -> 29.73 | Hurt |
| add pretraining to DRUNet backbone (ablation) | DRUNet + L_self · Poisson demosaicing BSDS500 (gamma=0.01) | PSNR (dB): 29.82 -> 30.00 | Improved |
| use pretrained CLIP decoder rather than untrained CLIP with supervised loss (ablation) | Frozen CLIP · Poisson demosaicing BSDS500 (gamma=0.01) | PSNR (dB): 21.91 -> 30.75 | Improved |
| use supervised loss instead of self-supervised loss | Frozen CLIP solver · Poisson demosaicing BSDS500 (gamma=0.01) | PSNR (dB): 30.53 -> 30.75 | Improved |
| replace GSPnP with Ours (Self) at severe noise | GSPnP · Poisson demosaicing BSDS500 (gamma=0.05) | PSNR (dB): 26.50 -> 26.98 | Improved |
| replace RAM with Ours (Self) at mild noise | RAM · Poisson demosaicing BSDS500 (gamma=0.01) | PSNR (dB): 30.46 -> 30.53 | Improved |
Closest papers by meaning
- Learning Cocoercive Conservative Denoisers via Helmholtz Decomposition for Poisson Imaging Inverse Problems (NeurIPS 2025 · similarity 0.87)
- Exact Posterior Score Estimation for Solving Linear Inverse Problems (NeurIPS 2026 · similarity 0.84)
- Self-diffusion for Solving Inverse Problems (NeurIPS 2025 · similarity 0.84)
- FiRe: Fixed-points of Restoration Priors for Solving Inverse Problems (CVPR 2025 · similarity 0.84)
- SP$^3$: Spherical Priors for Plug-and-Play Restoration (NeurIPS 2026 · similarity 0.83)
- Provably Robust Score-Based Diffusion Posterior Sampling for Plug-and-Play Image Reconstruction (NeurIPS 2024 · similarity 0.83)
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.
