Research map / Optimization / Optimizers
Self-Supervised Deep Image Restoration via Adaptive Stochastic Gradient Langevin Dynamics
Weixi Wang, Ji Li, Hui Ji
Check 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 adaptive noise perturbation instead of classic SGLD (ablation) | U-Net restoration trained with ASGLD · CS, Set11 and BSD68; tested noise and sampling ratios | PSNR (dB): 25.30 -> 26.02; 28.59 -> 29.35; 30.31 -> 31.11; 24.13 -> 24.56; 26.28 -> 27.16; 28.02 -> 28.75 | Improved |
| Use adaptive noise perturbation instead of classic SGLD (ablation) | U-Net restoration trained with ASGLD · CS-MRI, Gaussian 1D/2D and Radial masks | PSNR (dB): 25.11 -> 26.03; 26.25 -> 27.40; 26.30 -> 27.37 | Improved |
| Use adaptive noise perturbation instead of classic SGLD (ablation) | U-Net phase retrieval · Phase retrieval, AWGN and Poisson noise | PSNR (dB): 27.64 -> 28.42; 31.66 -> 32.42; 33.25 -> 34.82; 40.74 -> 42.04; 33.09 -> 35.85; 26.73 -> 29.15 | Improved |
| Use ASGLD rather than BNN | Self-supervised CS reconstruction · Set11 and BSD68, σ=10, 25% ratio | PSNR (dB): 28.67 -> 29.35; 26.47 -> 27.16 | Improved |
| Use ASGLD rather than DIP | Self-supervised CS-MRI reconstruction · 1D Gaussian mask, no noise | PSNR: 31.80 -> 32.18 | Improved |
| Use ASGLD rather than BNN | Self-supervised CS-MRI reconstruction · 2D Gaussian mask, σ=10% | PSNR: 27.12 -> 27.40 | Improved |
| Use ASGLD rather than the strongest listed baseline | Phase retrieval · Unnatural-6, AWGN SNR=15 | PSNR (dB): 32.81 -> 34.53 | Improved |
| Use ASGLD rather than the strongest listed baseline | Phase retrieval · Natural-6, AWGN SNR=15 | PSNR (dB): 29.59 -> 30.30 | Improved |
| Use ASGLD rather than the strongest listed baseline | Phase retrieval · Unnatural-6, Poisson α=9 | PSNR (dB): 39.41 -> 43.67 | Improved |
| Use ASGLD instead of SGLD (ablation) | MRI reconstruction · MRI reconstruction with measurement noise, σ=25.5 | elapsed time: ≈640s -> ≈190s | Improved |
| Use ASGLD instead of BNN | MRI reconstruction · MRI reconstruction with measurement noise, σ=25.5 | elapsed time: ≈520s -> ≈190s | Improved |
Closest papers by meaning
- Self-supervised Bayesian Deep Learning for Image Recovery with Applications to Compressive Sensing (ECCV 2020 · similarity 0.90)
- Stochastic Deep Restoration Priors for Imaging Inverse Problems (ICML 2025 · similarity 0.89)
- Analyzing the Sample Complexity of Self-Supervised Image Reconstruction Methods (NeurIPS 2023 · similarity 0.87)
- Stochastic Solutions for Linear Inverse Problems using the Prior Implicit in a Denoiser (NeurIPS 2021 · similarity 0.87)
- A Restoration Network as an Implicit Prior (ICLR 2024 · similarity 0.87)
- Plug-and-Play image restoration with Stochastic deNOising REgularization (ICML 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.
