Research map / Adversarial robustness and security / Certified and empirical robustness
Robustness of 3D Deep Learning in an Adversarial Setting
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 ISO adversarial occlusion instead of random occlusion | VoxNet · ModelNet40, 25% occlusion | classification accuracy: 60.9% -> 0% | Hurt |
| Use ISO adversarial occlusion instead of random occlusion | VoxNet · KITTI, 25% occlusion | classification accuracy: 55.5% -> 0% | Hurt |
| Use ISO adversarial occlusion instead of random occlusion | VoxNet · ModelNet10, 25% occlusion | classification accuracy: 72.1% -> 1.0% | Hurt |
| Use ISO adversarial occlusion instead of random occlusion | PointNet · ModelNet10, 25% occlusion | classification accuracy: 84.0% -> 27% | Hurt |
| Use ISO adversarial occlusion instead of random occlusion | PointNet · ModelNet40, 25% occlusion | classification accuracy: 79.4% -> 17.9% | Hurt |
| Use ISO adversarial occlusion instead of random occlusion | PointNet · KITTI, 25% occlusion | classification accuracy: 72.5% -> 49.5% | Hurt |
| Use ISO adversarial occlusion instead of random occlusion | PointNet · ModelNet40, 50% occlusion | classification accuracy: 78.2% -> 4.1% | Hurt |
| Use ISO adversarial occlusion instead of random occlusion | PointNet · KITTI, 75% occlusion | classification accuracy: 68.5% -> 0% | Hurt |
| Use ISO adversarial occlusion instead of random occlusion | VoxNet · ModelNet40, up to 6.5% occlusion | classification accuracy: 76.1% -> 0% | Hurt |
| Compute salience and iteratively remove critical points rather than remove points in random order | 3D deep-learning models · ModelNet10, ModelNet40, and KITTI | classification accuracy under occlusion: The paper reports random occlusion overestimates robustness by around 60%. | Hurt |
| Run ISO in black-box mode rather than white-box mode | PointNet and VoxNet · KITTI | classification accuracy at 25% occlusion: PointNet: 49.5% -> 49.5%; VoxNet: 0% -> 0% | No clear effect |
Closest papers by meaning
- Robust Adversarial Objects against Deep Learning Models (AAAI 2020 · similarity 0.89)
- Adversarially Robust 3D Point Cloud Recognition Using Self-Supervisions (NeurIPS 2021 · similarity 0.88)
- A Critical Revisit of Adversarial Robustness in 3D Point Cloud Recognition with Diffusion-Driven Purification (ICML 2023 · similarity 0.88)
- Exploring Architectural Ingredients of Adversarially Robust Deep Neural Networks (NeurIPS 2021 · similarity 0.88)
- Self-Robust 3D Point Recognition via Gather-Vector Guidance (CVPR 2020 · similarity 0.87)
- Interpreting Robustness Proofs of Deep Neural Networks (ICLR 2024 · similarity 0.86)
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.