Research map / Point clouds / Segmentation and detection
3D-VField: Adversarial Augmentation of Point Clouds for Domain Generalization in 3D Object Detection
Alexander Lehner, Stefano Gasperini, Alvaro Marcos-Ramiro, Michael Schmidt, Mohammad-Ali Nikouei Mahani, Nassir Navab, Benjamin Busam, Federico Tombari
Check 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 sensor-aware adversarial vector-field augmentation | PointPillars · Waymo val, transfer from KITTI | AP: 40.86 -> 44.61 | Improved |
| Add sensor-aware adversarial vector-field augmentation | PointPillars · CrashD rare crash, transfer from KITTI | AP: 22.48 -> 30.37 | Improved |
| Add sensor-aware adversarial vector-field augmentation | PointPillars · KITTI val, car, moderate | AP: 77.11 -> 77.13 | No clear effect |
| Add 3D-VField augmentation | Second · Waymo val, transfer from KITTI | AP: 42.45 -> 43.51 | Improved |
| Add 3D-VField augmentation | Part-A² · Waymo val, transfer from KITTI | AP: 49.76 -> 56.08 | Improved |
| Add statistical normalization domain adaptation to 3D-VField | PointPillars · CrashD rare crash, transfer from KITTI | AP: 48.23 -> 76.42 | Improved |
| Use 12 rather than 1 relative-rotation group (ablation) | 3D-VField · Waymo, transfer from KITTI | AP: 40.43 -> 44.61 | Improved |
| Use 360 rather than 12 relative-rotation groups (ablation) | 3D-VField · Waymo, transfer from KITTI | AP: 44.61 -> 40.30 | Hurt |
| Do not learn the perturbation; apply the constraints (ablation) | 3D-VField · Waymo, transfer from KITTI | AP: 41.62 vs 40.86 | No clear effect |
| Remove all deformation constraints (ablation) | 3D-VField · CrashD rare crash, transfer from KITTI | AP: 27.55 vs 30.37 | Hurt |
| Apply sensor-ray constraint without distance smoothing (ablation) | 3D-VField · CrashD rare crash, transfer from KITTI | AP: 29.16 vs 30.37 | Hurt |
| Increase maximum deformation from 30 cm to 40 cm or 60 cm (ablation) | 3D-VField · KITTI | AP: 1% and 1.7% | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PointPillars — PointPillars: Fast Encoders for Object Detection From Point Clouds (CVPR 2019) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- 3D Adversarial Augmentations for Robust Out-of-Domain Predictions (IJCV 2023 · similarity 0.89)
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · similarity 0.87)
- An Empirical Study of the Generalization Ability of Lidar 3D Object Detectors to Unseen Domains (CVPR 2024 · similarity 0.87)
- A New Adversarial Perspective for LiDAR-based 3D Object Detection (AAAI 2025 · similarity 0.87)
- Geometry-Guided Domain Generalization for Monocular 3D Object Detection (AAAI 2024 · similarity 0.87)
- DI-V2X: Learning Domain-Invariant Representation for Vehicle-Infrastructure Collaborative 3D Object Detection (AAAI 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.
