Research map / Image generation / Fast diffusion sampling
A Regularized Conditional GAN for Posterior Sampling in Image Recovery Problems
Matthew C Bendel, Rizwan Ahmad, Philip Schniter
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 supervised-l1 plus SD reward regularization instead of Ohayon et al. loss | conditional GAN · fastMRI R=4 | CFID (72 samples): 4.06 -> 3.10 | Improved |
| use proposed regularization instead of Ohayon et al. loss | conditional GAN · fastMRI R=8 | CFID (72 samples): 6.04 -> 4.87 | Improved |
| use proposed regularization instead of Ohayon et al. loss | conditional GAN · fastMRI R=4 | CFID (all 14576 samples): 2.95 -> 1.29 | Improved |
| use proposed cGAN instead of Adler & Oktem cGAN | conditional GAN · fastMRI R=8 | FID: 10.77 -> 7.72 | Improved |
| use proposed cGAN instead of Langevin sampling | conditional GAN · fastMRI R=8 | FID: 14.32 -> 7.72 | Improved |
| use proposed cGAN instead of Langevin sampling | conditional GAN · fastMRI | 4-sample generation time: 14 min -> 217 ms | Improved |
| compare APSD of Ohayon et al. cGAN with others | Ohayon et al. cGAN · fastMRI R=4 | APSD: 7.2e-8 vs 3.8e-6 | Hurt |
| use proposed regularization (and remove MBSD) instead of CoModGAN (ablation) | CoModGAN · CelebA-HQ inpainting | CFID: 5.29 -> 4.69 | Improved |
| use proposed regularization (and remove MBSD) instead of CoModGAN (ablation) | CoModGAN · CelebA-HQ inpainting | FID: 8.50 -> 7.45 | Improved |
| use proposed cGAN instead of Score-SDE | conditional GAN · CelebA-HQ inpainting | FID: 7.92 -> 7.45 | Improved |
| use proposed cGAN instead of Score-SDE | conditional GAN · CelebA-HQ inpainting | 128-sample generation time: 48 min -> 217 ms | Improved |
| increase number of averaged samples P from 1 to 32 | proposed cGAN · fastMRI R=4 | PSNR: 36.96 -> 39.55 | Improved |
Closest papers by meaning
- pcaGAN: Improving Posterior-Sampling cGANs via Principal Component Regularization (NeurIPS 2024 · similarity 0.90)
- Invertible generative models for inverse problems: mitigating representation error and dataset bias (ICML 2020 · similarity 0.87)
- Posterior Sampling Based on Gradient Flows of the MMD with Negative Distance Kernel (ICLR 2024 · similarity 0.86)
- UGoDIT: Unsupervised Group Deep Image Prior Via Transferable Weights (NeurIPS 2025 · similarity 0.86)
- Provably Robust Score-Based Diffusion Posterior Sampling for Plug-and-Play Image Reconstruction (NeurIPS 2024 · similarity 0.86)
- A Principled Self-Referenced Early Stopping Approach for Deep Image Prior (NeurIPS 2026 · 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.
