Research map / Detection and segmentation / CNN features
Real-World Deep Local Motion Deblurring
Haoying Li, Ziran Zhang, Tingting Jiang, Peng Luo, Huajun Feng, Zhihai Xu
This paper tried a common change: Pretrained initialization
Initialize with pretrained MIMO-Unet weights · ReLoBlur test · Weighted PSNR: 28.10 -> 28.32 Improved
What 86 accepted papers report for this change
- 74 improved
- 7 no clear effect
- 1 mixed
- 4 hurt
- Efficiently Training Time-to-First-Spike Spiking Neural Networks from Scratch ICML 2026 · initialize from pretrained ReLU weights · CIFAR10 · accuracy (%): 90.56 -> 93.89Improved
- Learning Visual Prior via Generative Pre-Training NeurIPS 2023 · initialize from pretrained GPT-2 text weights · COCO · KL divergence: 1.720 -> 1.959Hurt
- Mind-Omni: A Unified Multi-Task Framework for Brain-Vision-Language Modeling via Discrete Diffusion ICML 2026 · initialize from pre-trained Muddit instead of training from scratch · B→I&T description · BLEU1: 20.35 -> 29.12Improved
See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.
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 LBAG instead of MIMO-Unet | ReLoBlur deblurring · ReLoBlur test | Weighted PSNR: 27.95 -> 28.25 | Improved |
| Add pretrained MIMO-Unet backbone initialization | LBAG · ReLoBlur test | Weighted PSNR: 28.25 -> 28.32 | Improved |
| Add gate blocks that localize blur and gate reconstruction (ablation) | LBAG+ · ReLoBlur test | Weighted PSNR: 27.88 -> 28.32 | Improved |
| Use Blur-Aware Patch Cropping (ablation) | LBAG+ · ReLoBlur test | Weighted PSNR: 28.17 -> 28.32 | Improved |
| Add SSIM loss term (ablation) | LBAG+ · ReLoBlur test | Weighted SSIM: 0.8639 -> 0.8734 | Improved |
| Initialize with pretrained MIMO-Unet weights (ablation) | LBAG · ReLoBlur test | Weighted PSNR: 28.10 -> 28.32 | Improved |
| Train LBAG+ on real ReLoBlur data rather than synthetic local-blur data | LBAG+ · ReLoBlur test | Weighted PSNR: 27.42 -> 28.32 | Improved |
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.
- DeblurGAN-v2 — DeblurGAN: Blind Motion Deblurring Using Conditional Adversarial Networks (CVPR 2018) · better on 5 of 5 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
- OMoBlur: An Object Motion Blur Dataset and Benchmark for Real-World Local Motion Deblurring (CVPR 2026 · similarity 0.91)
- Adaptive Window Pruning for Efficient Local Motion Deblurring (ICLR 2024 · similarity 0.91)
- Parameterized Blur Kernel Prior Learning for Local Motion Deblurring (CVPR 2025 · similarity 0.90)
- GS-Blur: A 3D Scene-Based Dataset for Realistic Image Deblurring (NeurIPS 2024 · similarity 0.88)
- Real-World Efficient Blind Motion Deblurring via Blur Pixel Discretization (CVPR 2024 · similarity 0.87)
- Motion-adaptive Separable Collaborative Filters for Blind Motion Deblurring (CVPR 2024 · similarity 0.87)
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.
