Research map / Probabilistic and scientific modeling / Variational and generative sampling
Total Deep Variation for Linear Inverse Problems
Erich Kobler, Alexander Effland, Karl Kunisch, Thomas Pock
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 |
|---|---|---|---|
| Individually train TDV³ for each noise level instead of using the σ=25 model with rescaling | TDV denoising · Set12, BSDS68, and Urban100; σ=15, 25, 50 | PSNR: 33.01 vs 32.93; 27.59 vs 27.50 | Improved |
| Use TDV³ instead of FOCNet | Denoising · Set12; σ=15, 25, 50 | PSNR: 33.07 vs 33.01; 30.73 vs 30.66; 27.68 vs 27.59 | Hurt |
| Use TDV³ instead of FOCNet | Denoising · BSDS68; σ=15, 25, 50 | PSNR: 31.83 vs 31.82; 29.38 vs 29.37; 26.50 vs 26.45 | Hurt |
| Use TDV³ instead of FOCNet | Denoising · Urban100; σ=15, 25, 50 | PSNR: 33.15 vs 32.87; 30.64 vs 30.38; 27.40 vs 27.04 | Hurt |
| Use TDV³ instead of OISR-LF-s | Single-image super-resolution · BSDS100, ×2, ×3, and ×4 | PSNR: 32.18 vs 32.20; 28.98 vs 29.11; 27.50 vs 27.60 | No clear effect |
| Use TDV³ instead of OISR-LF-s | Single-image super-resolution · Set14, ×2, ×3, and ×4 | PSNR: 33.35 vs 33.62; 29.96 vs 30.35; 28.41 vs 28.63 | Hurt |
| Increase the number of TDV macro-blocks (ablation) | TDV denoising · BSDS68, σ=25 | PSNR | Improved |
| Increase discretization depth beyond S=10 (ablation) | TDV denoising · BSDS68, σ=25 | PSNR | No clear effect |
| Apply a denoising-trained TDV regularizer to CT reconstruction without retraining | TDV³₂₅ · MAYO CT, 4-fold and 8-fold angular undersampling | reconstruction quality | Improved |
| Apply a denoising-trained TDV regularizer to MRI reconstruction without retraining | TDV³₂₅ · MRI, 4-fold and 6-fold Cartesian undersampling | reconstruction quality | Improved |
Closest papers by meaning
- Weakly Convex Regularisers for Inverse Problems: Convergence of Critical Points and Primal-Dual Optimisation (ICML 2024 · similarity 0.88)
- A Restoration Network as an Implicit Prior (ICLR 2024 · similarity 0.87)
- Stochastic Deep Restoration Priors for Imaging Inverse Problems (ICML 2025 · similarity 0.87)
- DEALing with Image Reconstruction: Deep Attentive Least Squares (ICML 2025 · similarity 0.87)
- A Variational Perspective on Solving Inverse Problems with Diffusion Models (ICLR 2024 · similarity 0.87)
- Solving Inverse Problems via Diffusion Optimal Control (NeurIPS 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.
