Research map / Video / Video segmentation and tracking
Patch Craft: Video Denoising by Deep Modeling and Patch Matching
Gregory Vaksman, Michael Elad, Peyman Milanfar
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 |
|---|---|---|---|
| extend nearest neighbor search to 7 adjacent frames (T_s=3 instead of 0) (ablation) | S-CNN · DAVIS average over sigma 10-50 | PSNR: 33.63 -> 35.01 | Improved |
| extend nearest neighbor search to adjacent frames, sigma=10 (ablation) | S-CNN · DAVIS sigma=10 | PSNR: 38.38 -> 39.90 | Improved |
| extend nearest neighbor search to adjacent frames, sigma=50 (ablation) | S-CNN · DAVIS sigma=50 | PSNR: 30.51 -> 31.55 | Improved |
| add T-CNN temporal filtering after S-CNN (ablation) | PaCNet · DAVIS average over sigma 10-50 | PSNR: 35.01 -> 35.42 | Improved |
| add T-CNN temporal filtering, sigma=10 (ablation) | PaCNet · DAVIS sigma=10 | PSNR: 39.90 -> 39.97 | Improved |
| add T-CNN temporal filtering, sigma=50 (ablation) | PaCNet · DAVIS sigma=50 | PSNR: 31.55 -> 32.20 | Improved |
| patch-craft frames with CNN versus FastDVDnet (no explicit non-local similarity) | PaCNet · DAVIS average over sigma 10-50 | PSNR: 34.74 -> 35.42 | Improved |
| benefit versus FastDVDnet shrinks as noise grows | PaCNet · DAVIS sigma 10 vs 50 | PSNR: 38.93 -> 39.97; 31.90 -> 32.20 | Mixed |
| PaCNet versus FastDVDnet under clipped Gaussian noise | PaCNet · DAVIS clipped noise average | PSNR: 34.51 -> 35.54 | Improved |
| PaCNet versus FastDVDnet on Set8 | PaCNet · Set8 average | PSNR: 32.31 -> 32.68 | Improved |
| single-frame S-CNN-0 versus LIDIA, a similar NN-augmented image denoiser | S-CNN-0 · single image denoising average | PSNR: 31.11 -> 31.03 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FastDVDnet — FastDVDnet: Towards Real-Time Deep Video Denoising Without Flow Estimation (CVPR 2020) · better on 9 of 9 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Patch-Craft Self-Supervised Training for Correlated Image Denoising (CVPR 2023 · similarity 0.89)
- Efficient and Training-Free Single-Image Diffusion Models (CVPR 2026 · similarity 0.88)
- Temporal As a Plugin: Unsupervised Video Denoising with Pre-Trained Image Denoisers (ECCV 2024 · similarity 0.88)
- Patched Denoising Diffusion Models For High-Resolution Image Synthesis (ICLR 2024 · similarity 0.87)
- Classic Video Denoising in a Machine Learning World: Robust, Fast, and Controllable (CVPR 2025 · similarity 0.87)
- Hierarchical Patch Diffusion Models for High-Resolution Video Generation (CVPR 2024 · similarity 0.86)
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.
