Research map / Image generation / Image restoration
Deep Burst Denoising
Clement Godard, Kevin Matzen, Matt Uyttendaele
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 |
|---|---|---|---|
| use recurrent two-track multi-frame network instead of concatenating 8 frames (ablation) | Concat C8F · Live Photos ablation, AWGN sigma=50, 8-frame training | PSNR (dB): 32.15 -> 33.62 | Improved |
| use recurrent network instead of concatenating 2 frames (ablation) | Concat C2F · Live Photos ablation, AWGN sigma=50, 8-frame training | PSNR (dB): 30.89 -> 33.62 | Improved |
| add stabilization of burst before denoising (ablation) | Ours · Live Photos ablation, AWGN sigma=50, 8-frame training | PSNR (dB): 32.60 -> 33.62 | Improved |
| increase depth from 8 to 12 layers (ablation) | Ours 8L · Live Photos ablation, AWGN sigma=50, 8-frame training | PSNR (dB): 33.62 -> 33.80 | Improved |
| increase depth from 8 to 16 layers (ablation) | Ours 8L · Live Photos ablation, AWGN sigma=50, 8-frame training | PSNR (dB): 33.62 -> 33.35 | Hurt |
| increase depth from 8 to 20 layers (ablation) | Ours 8L · Live Photos ablation, AWGN sigma=50, 8-frame training | PSNR (dB): 33.62 -> 33.48 | Hurt |
| reduce depth from 8 to 4 layers (ablation) | Ours 8L · Live Photos ablation, AWGN sigma=50, 8-frame training | PSNR (dB): 33.62 -> 33.01 | Hurt |
| pre-train single-frame track before multi-frame training (ablation) | Ours MFD · Live Photos, AWGN sigma=50 | PSNR (dB): 0.5dB lead | Improved |
| train on 8-frame sequences instead of 2 or 4 frames (ablation) | Ours MFD · Live Photos, 16-frame test | generalization to longer sequences | Improved |
| feed sequence forward then backward at test time (ablation) | Ours MFD · Live Photos | burn-in and flickering | Improved |
| use multi-frame recurrent network instead of single-frame DenoiseNet, sigma=50 | DenoiseNet · Live Photos multi-frame (16-frame test seqs), AWGN sigma=50 | PSNR (dB): 29.56 -> 33.62 | Improved |
| use multi-frame network instead of VBM4D, sigma=75 | VBM4D · Live Photos multi-frame (16-frame test seqs), AWGN sigma=75 | PSNR (dB): 26.60 -> 31.44 | Improved |
Closest papers by meaning
- Burst Denoising via Temporally Shifted Wavelet Transforms (ECCV 2020 · similarity 0.91)
- Iterative Residual CNNs for Burst Photography Applications (CVPR 2019 · similarity 0.91)
- Efficient Burst Raw Denoising with Variance Stabilization and Multi-frequency Denoising Network (IJCV 2022 · similarity 0.90)
- NAN: Noise-Aware NeRFs for Burst-Denoising (CVPR 2022 · similarity 0.89)
- Contrastive Learning for Low-Light Raw Denoising (Student Abstract) (AAAI 2024 · similarity 0.86)
- Real-World Mobile Image Denoising Dataset with Efficient Baselines (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.
