Research map / Image generation / Image restoration
Zero-Order Reverse Filtering
Xin Tao, Chao Zhou, Xiaoyong Shen, Jue Wang, Jiaya Jia
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 |
|---|---|---|---|
| apply zero-order reverse filtering iterations to Gaussian-filtered images | filtered image · BSD300-based, GS | GT PSNR (dB): 27.75 -> 41.70 | Improved |
| apply zero-order reverse filtering to bilateral-filtered images | filtered image · BSD300-based, BF | GT PSNR (dB): 25.50 -> 45.28 | Improved |
| apply zero-order reverse filtering to guided-filtered images | filtered image · BSD300-based, GF | GT PSNR (dB): 28.21 -> 51.05 | Improved |
| apply zero-order reverse filtering to median-filtered images | filtered image · BSD300-based, MF | GT PSNR (dB): 26.01 -> 26.07 | No clear effect |
| apply zero-order reverse filtering to weighted median filtered images | filtered image · BSD300-based, WMF | GT PSNR (dB): 33.00 -> 22.98 (final), 35.43 (best) | Mixed |
| apply zero-order reverse filtering to local extrema filtered images | filtered image · BSD300-based, LE | GT PSNR (dB): 29.30 -> 7.96 (final), 44.21 (best) | Mixed |
| run 50 iterations instead of stopping at best iteration for partially reversible bilateral grid (ablation) | zero-order reverse filtering · BSD300-based, BFG | GT PSNR (dB): 37.70 -> 35.78 | Hurt |
| run 50 iterations instead of stopping at best iteration for domain transform filter (ablation) | zero-order reverse filtering · BSD300-based, RF | GT PSNR (dB): 31.16 -> 28.63 | Hurt |
| run 50 iterations instead of stopping at best iteration for permutohedral lattice (ablation) | zero-order reverse filtering · BSD300-based, BFPL | GT PSNR (dB): 36.92 -> 32.97 | Hurt |
| add reverse filtering to initialization with bicubic upsampling | bicubic x2 super-resolution · single example image | PSNR (dB): 28.99 -> 30.50 | Improved |
| use zero-order reverse filtering instead of Lanczos3 interpolation | x2 super-resolution · single example image | PSNR (dB): 29.57 -> 30.50 | Improved |
| add reverse filtering as post-processing to ScSR and SRCNN outputs | learning-based super-resolution · single example image | PSNR | Improved |
Closest papers by meaning
- UniDemoiré: Towards Universal Image Demoiréing with Data Generation and Synthesis (AAAI 2025 · similarity 0.84)
- DiracDiffusion: Denoising and Incremental Reconstruction with Assured Data-Consistency (ICML 2024 · similarity 0.83)
- Learning Cocoercive Conservative Denoisers via Helmholtz Decomposition for Poisson Imaging Inverse Problems (NeurIPS 2025 · similarity 0.83)
- Learning Pseudo-Contractive Denoisers for Inverse Problems (ICML 2024 · similarity 0.82)
- MAP Estimation with Denoisers: Convergence Rates and Guarantees (NeurIPS 2025 · similarity 0.82)
- Beyond First-Order Tweedie: Solving Inverse Problems using Latent Diffusion (CVPR 2024 · similarity 0.82)
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.
