Research map / Adversarial robustness and security / Attacks on text and graphs
Physical Attack on Monocular Depth Estimation with Optimal Adversarial Patches
Zhiyuan Cheng, James Liang, Hongjun Choi, Guanhong Tao, Zhiwen Cao, Dongfang Liu, Xiangyu Zhang
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 |
|---|---|---|---|
| add object-oriented adversarial loss (no regional optimization) (ablation) | patch attack on Monodepth2 · vehicle, KITTI | mean depth estimation error Ed: 8.47 -> 14.95 | Improved |
| add regional optimization without object-oriented loss (ablation) | patch attack on Monodepth2 · vehicle, KITTI | mean depth estimation error Ed: 8.47 -> 6.38 | Hurt |
| add regional optimization on top of object-oriented loss (ablation) | patch attack on Monodepth2 · vehicle, KITTI | mean depth estimation error Ed: 14.95 -> 16.84 | Improved |
| add regional optimization on top of object-oriented loss (ablation) | patch attack on Monodepth2 · vehicle, KITTI | ratio of affected region Ra: 0.52 -> 0.65 | Improved |
| replace location-optimized patch with sensitive region localization | patch attack on Monodepth2 · vehicle, KITTI | mean depth estimation error Ed: 13.90 -> 16.84 | Improved |
| replace location-optimized patch with sensitive region localization | patch attack on DepthHints · vehicle, KITTI | mean depth estimation error Ed: 2.51 -> 15.23 | Improved |
| replace location-optimized patch with sensitive region localization | patch attack on Manydepth · vehicle, KITTI | mean depth estimation error Ed: 3.03 -> 6.31 | Improved |
| raise style transfer weight from 1 to 10 (ablation) | patch attack on Monodepth2 · vehicle, KITTI | mean depth estimation error Ed: 7.69 -> 1.44 | Hurt |
| lower style transfer weight from 1 to 0.1 (ablation) | patch attack on Monodepth2 · vehicle, KITTI | mean depth estimation error Ed: 7.69 -> 11.09 | Improved |
| raise style transfer weight from 1 to 100 (ablation) | patch attack on Monodepth2 · vehicle, KITTI | SSIM to style image: 0.575 -> 0.993 | Improved |
| enlarge patch from 1/9 to 1/3 of rear area (ablation) | patch attack on Monodepth2 · PointPillars 3D detection | attack success rate: 95% -> 100% | Improved |
| enlarge patch from 1/9 to 2/9 of rear area (ablation) | patch attack on DepthHints · PointPillars 3D detection | attack success rate: 93% -> 97% | Improved |
Closest papers by meaning
- Beware of Road Markings: A New Adversarial Patch Attack to Monocular Depth Estimation (NeurIPS 2024 · similarity 0.93)
- Physical 3D Adversarial Attacks against Monocular Depth Estimation in Autonomous Driving (CVPR 2024 · similarity 0.92)
- DepthVanish: Optimizing Adversarial Interval Structures for Stereo-Depth-Invisible Patches (NeurIPS 2025 · similarity 0.89)
- Fusion Is Not Enough: Single Modal Attacks on Fusion Models for 3D Object Detection (ICLR 2024 · similarity 0.86)
- BadPart: Unified Black-box Adversarial Patch Attacks against Pixel-wise Regression Tasks (ICML 2024 · similarity 0.86)
- Seeing is Not Believing: Adversarial Natural Object Optimization for Hard-Label 3D Scene Attacks (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.
