Research map / Adversarial robustness and security / Attacks on text and graphs
Distracting Downpour: Adversarial Weather Attacks for Motion Estimation
Jenny Schmalfuss, Lukas Mehl, Andrés Bruhn
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 |
|---|---|---|---|
| augment 50% of Sintel-final training frames with random snow when training RAFT (ablation) | RAFT · Sintel snow augmentation | robustness AEE: 4.19 -> 0.91 | Improved |
| augment 50% of Sintel-final training frames with random snow (ablation) | RAFT · Sintel test clean | EPE: 1.642 -> 1.589 | Improved |
| augment 50% of training frames with snow (ablation) | RAFT · KITTI train | F1-all: 5.65 -> 5.54 | Improved |
| augment 100% instead of 50% of training frames with snow (ablation) | RAFT · Sintel test final | EPE: 3.155 -> 3.384 | Hurt |
| snow augmentation in training (ablation) | RAFT · fog attack | robustness AEE: 0.87 -> 1.04 | Hurt |
| increase number of weather particles from 1000 to 5000 (ablation) | optical flow methods · Sintel train random augmentation | robustness AEE: 3.94 -> 23.42 (FlowNet2) | Hurt |
| adversarially optimize particle parameters instead of random placement (ablation) | GMA · Sintel-tr115 | robustness AEE: 3.77 -> 9.50 | Hurt |
| optimize motion offset parameters before motion only vs transparency only (ablation) | particle attack · Sintel-tr115 | robustness AEE: RAFT 7.39 vs 5.29; FlowFormer 5.33 vs 3.56 | Mixed |
| use adversarial snow rather than L_p attacks on high-quality methods | RAFT, GMA, FlowFormer · Sintel train | robustness AEE: FlowFormer PCFA 14.68 vs snow 7.10 | Mixed |
| use alpha blending instead of additive blending for color (ablation) | particle augmentation · Sintel train | robustness AEE: white FN2 14.05 -> 10.03 | Hurt |
Closest papers by meaning
- Towards Robust Rain Removal Against Adversarial Attacks: A Comprehensive Benchmark Analysis and Beyond (CVPR 2022 · similarity 0.88)
- Let it Snow! Animating 3D Gaussian Scenes with Dynamic Weather Effects via Physics-Guided Score Distillation (CVPR 2026 · similarity 0.85)
- All in One Bad Weather Removal Using Architectural Search (CVPR 2020 · similarity 0.84)
- Recurrent Multi-Frame Deraining: Combining Physics Guidance and Adversarial Learning (TPAMI 2021 · similarity 0.84)
- WeatherEdit: Controllable Weather Editing with 4D Gaussian Field (AAAI 2026 · similarity 0.84)
- Hide in Thicket: Generating Imperceptible and Rational Adversarial Perturbations on 3D Point Clouds (CVPR 2024 · similarity 0.84)
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.
