Research map / Adversarial robustness and security / Attacks on text and graphs
Watch out! Motion is Blurring the Vision of Your Deep Neural Networks
Qing Guo, Felix Juefei-Xu, Xiaofei Xie, Lei Ma, Jian Wang, Bing Yu, Wei Feng, Yang Liu
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 saliency-regularized motion blur with adaptively tuned object and background translations instead of pixel-wise independent kernels (ablation) | ABBA · NeurIPS'17 adversarial competition dataset, Inc-v3 | success rate: 89.2 -> 65.6 | Hurt |
| Add adaptive joint object/background translations rather than using shared image-wide kernels and translations (ablation) | ABBA · NeurIPS'17 adversarial competition dataset, Inc-v3 | success rate: 62.4 -> 65.6 | Improved |
| Restrict motion blur to the salient object (ablation) | ABBA · NeurIPS'17 adversarial competition dataset, Inc-v3 | success rate: 65.6 -> 21.0 | Hurt |
| Restrict motion blur to the background (ablation) | ABBA · NeurIPS'17 adversarial competition dataset, Inc-v3 | success rate: 65.6 -> 30.9 | Hurt |
| Increase the valid kernel size and maximum translation (ablation) | ABBA · NeurIPS'17 adversarial competition dataset | success rate | Improved |
| Set the object motion direction near 45° (ablation) | ABBA · NeurIPS'17 adversarial competition dataset | success rate | Improved |
| Use average kernels and shared object/background translation for physical camera motion instead of learned kernels and separate translations (ablation) | ABBA_physical · AirSim cars, Inc-v3 | success rate: 97.6 -> 85.3 | Hurt |
| Use physical camera motion instead of the digitally generated adversarial blur (ablation) | ABBA · AirSim cars, Inc-v4 transfer | success rate: 87.8 -> 9.7 | Hurt |
| Use ABBA rather than TIDIM | Inc-v3-generated attacks · NeurIPS'17 adversarial competition dataset, Inc-v3 ens3 | success rate: 46.9 vs 46.6 | No clear effect |
| Use ABBA rather than TIDIM | Inc-v3-generated attacks · NeurIPS'17 adversarial competition dataset, HGD | success rate: 38.3 vs 31.0 | Hurt |
Closest papers by meaning
- Learning To Adversarially Blur Visual Object Tracking (ICCV 2021 · similarity 0.90)
- Blur-Invariant Deep Learning for Blind-Deblurring (ICCV 2017 · similarity 0.87)
- From Motion Blur to Motion Flow: A Deep Learning Solution for Removing Heterogeneous Motion Blur (CVPR 2017 · similarity 0.87)
- DeblurGAN: Blind Motion Deblurring Using Conditional Adversarial Networks (CVPR 2018 · similarity 0.86)
- Non-Blind Deblurring: Handling Kernel Uncertainty With CNNs (CVPR 2018 · similarity 0.86)
- DynaMoDe-NeRF: Motion-aware Deblurring Neural Radiance Field for Dynamic Scenes (CVPR 2025 · similarity 0.85)
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.
