Research map / Point clouds / Segmentation and detection
PseudoAugment: Learning to Use Unlabeled Data for Data Augmentation in Point Clouds
Zhaoqi Leng, Shuyang Cheng, Benjamin Caine, Weiyue Wang, Xiao Zhang, Jonathon Shlens, Mingxing Tan, Dragomir Anguelov
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 PseudoFrame (drop low-confidence boxes and their points) instead of plain pseudo labeling | Pseudo labeling, PointPillars · Waymo val Vehicle 10% | L1 AP: 50.7 -> 51.1 | Improved |
| add PseudoBBox and PseudoBackground on top of PseudoFrame (ablation) | PseudoFrame, PointPillars · Waymo val 10% | AP: +3.2 on Vehicle AP and + 1.2 on Pedestrian AP | Improved |
| replace pseudo labeling with all PseudoAugments | PointPillars student · Waymo val Vehicle 10% | L1 AP: 50.7 -> 54.3 | Improved |
| replace pseudo labeling with all PseudoAugments | PointPillars student · Waymo val Pedestrian 10% | L1 AP: 56.7 -> 58.4 | Improved |
| use PseudoAugments instead of pseudo labeling for a larger Pillars2X student | Pillars2X · Waymo val Vehicle 10% | AP: 51.6 -> 55.7 | Improved |
| use PseudoAugments instead of pseudo labeling when distilling to point-based StarNet | StarNet · Waymo val Vehicle 10% | AP: 48.2 -> 51.2 | Improved |
| use PseudoAugments instead of pseudo labeling on KITTI (moderate vehicle) | PointPillars · KITTI val Vehicle moderate 10% | AP: 51.4 -> 56.5 | Improved |
| use AutoPseudoAugment instead of PPBA at 10% labeled | PointPillars · Waymo val Vehicle 10% | AP L1: 50.2 -> 56.7 | Improved |
| use AutoPseudoAugment instead of pseudo labeling at 50% labeled | PointPillars · Waymo val Pedestrian 50% | AP L1: 68.2 -> 69.6 | Improved |
| use pseudo labeling instead of PPBA at 50% labeled | PointPillars · Waymo val Vehicle 50% | AP L1: 60.9 -> 59.8 | Hurt |
| add PseudoBBox instead of ground-truth box pasting (GTBBox) (ablation) | PointPillars · Waymo val Vehicle 10% | L1 AP: 44.7 -> 46.4 | Improved |
| add PseudoBackground versus PseudoBBox for pedestrians (ablation) | PointPillars · Waymo val Pedestrian 10% | L1 AP: PseudoBackground +3.1 AP vs PseudoBBox +1.2 AP | Improved |
Closest papers by meaning
- Pseudo Label Refinery for Unsupervised Domain Adaptation on Cross-dataset 3D Object Detection (CVPR 2024 · similarity 0.88)
- Fully Data-Driven Pseudo Label Estimation for Pointly-Supervised Panoptic Segmentation (AAAI 2024 · similarity 0.87)
- Leveraging Imagery Data with Spatial Point Prior for Weakly Semi-supervised 3D Object Detection (AAAI 2024 · similarity 0.87)
- Mining and Transferring Feature-Geometry Coherence for Unsupervised Point Cloud Registration (NeurIPS 2024 · similarity 0.87)
- Dual-Perspective Knowledge Enrichment for Semi-supervised 3D Object Detection (AAAI 2024 · similarity 0.86)
- Semi-supervised 3D Object Detection with PatchTeacher and PillarMix (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.
