Research map / Image generation / Image restoration
Image Restoration via Primal Dual Hybrid Gradient and Flow Generative Model
Ji Li, Chao Wang
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 l1 fidelity instead of squared l2 with PDHG (ablation) | PDHG flow PnP · AFHQ-Cat impulse noise, denoising | PSNR: 25.81 -> 37.94 | Improved |
| use l1 fidelity instead of l2 under Poisson noise (ablation) | PDHG flow PnP · AFHQ-Cat Poisson noise, denoising | PSNR: 34.82 -> 29.31 | Hurt |
| replace squared l2 fidelity (PnP-FBS) with PDHG and l1 fidelity for impulse noise | flow-based PnP · FFHQ salt-and-pepper denoising | PSNR: 26.27 -> 40.27 | Improved |
| replace squared l2 fidelity with l2-norm fidelity for Poisson noise | flow-based PnP · FFHQ Poisson super-resolution | PSNR: 23.51 -> 29.96 | Improved |
| use PDHG fidelity framework under Gaussian noise | flow-based PnP · CelebA Gaussian noise | PSNR: 31.95 -> 31.40 (denoising) | No clear effect |
| use PnP-FBS gradient step rather than exact proximal step (PnP-HQS) | squared-l2 PnP methods · FFHQ Poisson deblurring | PSNR: 23.25 -> 28.45 | Improved |
| use one denoiser sample instead of five and avoid sequential approximation | PDHG flow PnP · time to deblur one image | seconds: 25 -> 13 (vs PnP-FBS) | Improved |
Closest papers by meaning
- PnP-Flow: Plug-and-Play Image Restoration with Flow Matching (ICLR 2025 · similarity 0.93)
- Plug-and-Play image restoration with Stochastic deNOising REgularization (ICML 2024 · similarity 0.91)
- Reconciling Stochastic and Deterministic Strategies for Zero-shot Image Restoration using Diffusion Model in Dual (CVPR 2025 · similarity 0.90)
- Provably Robust Score-Based Diffusion Posterior Sampling for Plug-and-Play Image Reconstruction (NeurIPS 2024 · similarity 0.90)
- Solving Inverse Problems with FLAIR (NeurIPS 2025 · similarity 0.90)
- Learning Cocoercive Conservative Denoisers via Helmholtz Decomposition for Poisson Imaging Inverse Problems (NeurIPS 2025 · similarity 0.89)
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.
