Research map / Image generation / Image restoration
Event-guided Multi-patch Network with Self-supervision for Non-uniform Motion Deblurring
Hongguang Zhang, Limeng Zhang, Yuchao Dai, Hongdong Li, Piotr Koniusz
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 |
|---|---|---|---|
| stack multiple MPN units (StackMPN) | MPN(1-2-4) · GoPro | PSNR: 30.21 -> 31.16 | Improved |
| add event input (E-MPN) | MPN · GoPro | PSNR: 30.21 -> 33.14 | Improved |
| add event input to stacked model | StackMPN · GoPro | PSNR: 31.16 -> 33.56 | Improved |
| use 5 blurry frames as input | MPN · GoPro | PSNR: 30.21 -> 30.89 | Improved |
| use 7 frames instead of 5 (ablation) | MPN · GoPro | PSNR: 30.89 -> 30.58 | Hurt |
| add finer patch levels beyond 1-2-4-8 (5 levels) (ablation) | MPN · GoPro | PSNR: 30.25 -> 29.87 | Hurt |
| use multi-patch hierarchy instead of multi-scale at 3 levels (ablation) | MPN(1-2-4) vs MSN(3) · GoPro | PSNR: 28.97 -> 30.21 | Improved |
| add multi-scale network on top of MPN (ablation) | MPN(1-2-4) · GoPro | PSNR / runtime: 30.21 -> 30.34; 17 ms -> 523 ms | No clear effect |
| share weights across levels (ablation) | MPN(1-2-4) · GoPro | PSNR: 30.21 -> 29.56 | Hurt |
| use event representation instead of TV-L1 optical flow (ablation) | E-StackMPN · VideoDeblurring | PSNR: 32.01 -> 32.57 | Improved |
| add Gaussian-noise self-supervision (ablation) | MPN · GoPro with Gaussian noise | PSNR: 20.98 -> 30.31 | Improved |
| add rotation self-supervision (ablation) | MPN · GoPro | PSNR: 30.24 -> 30.85 | Improved |
Closest papers by meaning
- Deep Stacked Hierarchical Multi-Patch Network for Image Deblurring (CVPR 2019 · similarity 0.92)
- Generalizing Event-Based Motion Deblurring in Real-World Scenarios (ICCV 2023 · similarity 0.91)
- Motion Deblurring With Real Events (ICCV 2021 · similarity 0.90)
- Learning Scale-Aware Spatio-temporal Implicit Representation for Event-based Motion Deblurring (ICML 2024 · similarity 0.88)
- Motion-adaptive Transformer for Event-based Image Deblurring (AAAI 2025 · similarity 0.88)
- Frequency-aware Event-based Video Deblurring for Real-World Motion Blur (CVPR 2024 · similarity 0.88)
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.
