Research map / Adversarial robustness and security / Black-box and transfer attacks
A Perturbation-Constrained Adversarial Attack for Evaluating the Robustness of Optical Flow
Jenny Schmalfuss, Philipp Scholze, 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 |
|---|---|---|---|
| use change of variables instead of clipping for the box constraint (ablation) | PCFA on RAFT · KITTI test, zero-flow target | AEE to target: 3.76 -> 3.54 | Improved |
| use change of variables instead of clipping for the box constraint (ablation) | PCFA on RAFT · KITTI test, negative-flow target | AEE to target: 22.48 -> 18.84 | Improved |
| use AEE loss instead of cosine similarity loss (ablation) | PCFA on RAFT · KITTI test, negative-flow target | AEE to target: 86.00 -> 18.84 | Improved |
| use AEE loss instead of cosine similarity loss (ablation) | PCFA on RAFT · KITTI test, zero-flow target | AEE to target: 32.37 -> 3.54 | Improved |
| use AEE loss instead of MSE loss (ablation) | PCFA on RAFT · KITTI test, zero-flow target | AEE to target: 7.46 -> 3.54 | Improved |
| use PCFA instead of I-FGSM | global attack · KITTI zero-flow, all networks | AEE to target | Improved |
| use disjoint instead of joint perturbation (frame-specific) (ablation) | PCFA on RAFT · KITTI test | AEE to target: 5.35 -> 3.76 | Improved |
| use joint instead of disjoint perturbation (universal) (ablation) | PCFA on RAFT · KITTI test | AEE to target: 28.88 -> 28.53 | Improved |
| use PCFA global perturbation instead of Patch Attack | universal perturbation on RAFT · KITTI test | AEE(f_adv,f): 0.76 -> 3.52 | Improved |
| use PCFA instead of Patch Attack | universal perturbation on GMA · KITTI test | AEE(f_adv,f): 0.95 -> 3.81 | Improved |
| transfer RAFT perturbation to FlowNet2 instead of white-box | universal joint PCFA · KITTI test | AEE(f_adv,f): 3.52 -> 2.88 | Hurt |
| transfer SpyNet perturbation to RAFT instead of white-box on SpyNet | universal joint PCFA · KITTI test | AEE(f_adv,f): 2.25 -> 1.87 | Mixed |
Closest papers by meaning
- Attacking Optical Flow (ICCV 2019 · similarity 0.90)
- Breaking Barriers in Physical-World Adversarial Examples: Improving Robustness and Transferability via Robust Feature (AAAI 2025 · similarity 0.86)
- Stop Walking in Circles! Bailing Out Early in Projected Gradient Descent (CVPR 2025 · similarity 0.85)
- Towards Million-Scale Adversarial Robustness Evaluation With Stronger Individual Attacks (CVPR 2025 · similarity 0.85)
- Advancing Adversarial Robustness in GNeRFs: The IL2-NeRF Attack (CVPR 2025 · similarity 0.85)
- Pixel2Feature Attack (P2FA): Rethinking the Perturbed Space to Enhance Adversarial Transferability (ICML 2025 · 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.
