Research map / Image generation / Image restoration
Trainable Nonlinear Reaction Diffusion: A Flexible Framework for Fast and Effective Image Restoration
Yunjin Chen, Thomas Pock
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 |
|---|---|---|---|
| train the influence functions in addition to filters (shared stage parameters) (ablation) | TNRD 5x5 10 stages · BSD68 denoising (sigma=25, 10-stage 5x5 ablation) | PSNR (dB): 28.47 -> 28.60 | Improved |
| train the influence functions in addition to filters (stage-varying parameters) (ablation) | TNRD 5x5 10 stages · BSD68 denoising (sigma=25, 10-stage 5x5 ablation) | PSNR (dB): 28.56 -> 28.86 | Improved |
| let parameters vary across stages (filters only) (ablation) | TNRD 5x5 10 stages · BSD68 denoising (sigma=25, 10-stage 5x5 ablation) | PSNR (dB): 28.47 -> 28.56 | Improved |
| let parameters vary across stages (filters and influence functions) (ablation) | TNRD 5x5 10 stages · BSD68 denoising (sigma=25, 10-stage 5x5 ablation) | PSNR (dB): 28.60 -> 28.86 | Improved |
| increase stages from 2 to 5 (ablation) | TNRD 5x5 · BSD68 denoising (sigma=25) | PSNR (dB): 28.58 -> 28.78 | Improved |
| increase stages from 5 to 8 (ablation) | TNRD 5x5 · BSD68 denoising (sigma=25) | PSNR (dB): 28.78 -> 28.83 | No clear effect |
| use trained TNRD instead of CSF of same capacity | TNRD 5x5 5 stages · BSD68 denoising (sigma=25) | PSNR (dB): 28.60 -> 28.78 | Improved |
| joint training with random initialization instead of greedy initialization (ablation) | TNRD 5x5 5 stages · BSD68 denoising (sigma=25, initialization) | PSNR (dB): 28.78 -> 28.61 | Hurt |
| plain initialization instead of greedy-trained initialization (ablation) | TNRD 5x5 5 stages · BSD68 denoising (sigma=25, initialization) | PSNR (dB): 28.78 -> 28.75 | No clear effect |
| plain initialization instead of greedy-trained initialization (ablation) | TNRD 7x7 5 stages · BSD68 denoising (sigma=25) | PSNR (dB): 28.92 -> 28.91 | No clear effect |
| increase filter size from 5x5 to 7x7 (ablation) | TNRD 5 stages · BSD68 denoising (sigma=25) | PSNR (dB): +0.14dB | Improved |
| increase filter size from 7x7 to 9x9 (ablation) | TNRD 5 stages · BSD68 denoising (sigma=25) | PSNR (dB): 28.96dB, +0.05dB | No clear effect |
Closest papers by meaning
- Improving Diffusion-Based Image Restoration with Error Contraction and Error Correction (AAAI 2024 · similarity 0.90)
- A Variational Perspective on Solving Inverse Problems with Diffusion Models (ICLR 2024 · similarity 0.89)
- A Modular Conditional Diffusion Framework for Image Reconstruction (NeurIPS 2024 · similarity 0.89)
- Residual Denoising Diffusion Models (CVPR 2024 · similarity 0.89)
- Reconciling Stochastic and Deterministic Strategies for Zero-shot Image Restoration using Diffusion Model in Dual (CVPR 2025 · similarity 0.89)
- Solving Video Inverse Problems Using Image Diffusion Models (ICLR 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.
