Research map / Detection and segmentation / 3D detection and driving
Improving 3D Object Detection through Progressive Population Based Augmentation
Shuyang Cheng, Zhaoqi Leng, Ekin Dogus Cubuk, Barret Zoph, Chunyan Bai, Jiquan Ngiam, Yang Song, Benjamin Caine, Vijay Vasudevan, Congcong Li, Quoc V. Le, Jonathon Shlens and 1 more
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 PPBA-learned augmentation | StarNet · KITTI test (Car, moderate) | 3D mAP: 73.99 -> 77.65 | Improved |
| add PPBA-learned augmentation | PointPillars · Waymo Open test (vehicle, Level 1, overall) | 3D mAP: 63.3 -> 67.5 | Improved |
| add PPBA-learned augmentation | PointPillars · Waymo Open test (pedestrian, Level 1, overall) | 3D mAP: 62.1 -> 66.4 | Improved |
| add PPBA-learned augmentation | StarNet · Waymo Open test (vehicle, Level 1, overall) | 3D mAP: 61.5 -> 64.6 | Improved |
| search progressively over a subset of operations with historical best parameters instead of full-space PBA (ablation) | PPBA · KITTI test (StarNet, Pedestrian moderate) | 3D mAP: 34.48 -> 44.08 | Improved |
| search progressively over a subset of operations instead of full-space PBA (ablation) | PPBA · KITTI test (StarNet, Car moderate) | 3D mAP: 75.02 -> 77.65 | Improved |
| use PPBA instead of random search over 1,000 policies | StarNet · KITTI test (Car, moderate) | 3D mAP: 74.94 -> 77.65 | Improved |
| use PPBA instead of random search on pedestrian | StarNet · KITTI test (Pedestrian, moderate) | 3D mAP: 44.71 -> 44.08 | No clear effect |
| add PPBA with 10% of training frames | PointPillars · Waymo Open val (10% single frames, Car) | 3D mAP: 52.4 -> 59.8 | Improved |
| add PPBA with 10% of training run segments | PointPillars · Waymo Open val (10% run segments, Car) | 3D mAP: 42.5 -> 54.2 | Improved |
| add a random augmentation policy from the search space (ablation) | PointPillars · Waymo Open val (10% single frames, Car) | 3D mAP: 52.4 -> 58.3 | Improved |
| add PPBA instead of random policy at 10% run segments (ablation) | PointPillars · Waymo Open val (10% run segments, Car) | 3D mAP: 49.5 -> 54.2 | Improved |
Closest papers by meaning
- PseudoAugment: Learning to Use Unlabeled Data for Data Augmentation in Point Clouds (ECCV 2022 · similarity 0.88)
- AGO-Net: Association-Guided 3D Point Cloud Object Detection Network (TPAMI 2021 · similarity 0.87)
- Domain Generalization of 3D Object Detection by Density-Resampling (ECCV 2024 · similarity 0.86)
- Structure Aware Single-Stage 3D Object Detection From Point Cloud (CVPR 2020 · similarity 0.86)
- LION: Linear Group RNN for 3D Object Detection in Point Clouds (NeurIPS 2024 · similarity 0.86)
- STONE: A Submodular Optimization Framework for Active 3D Object Detection (NeurIPS 2024 · 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.
