Research map / Point clouds / Segmentation and detection
PointAugmenting: Cross-Modal Augmentation for 3D Object Detection
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 |
|---|---|---|---|
| Add point-wise detection CNN features and late fusion to LiDAR detection | CenterPoint · nuScenes test | mAP: 60.3 -> 66.8 | Helped |
| Add camera features with cross-modal fusion | CenterPoint · Waymo validation, L2 | mAP: 66.88 -> 69.22 | Helped |
| Use detection CNN features instead of segmentation scores (ablation) | PointPainting-style point decoration · nuScenes 1/8 validation | mAP: 42.3 -> 46.0 | Helped |
| Use late BEV fusion instead of point-wise concatenation (ablation) | Detection features + point-wise concatenation · nuScenes 1/8 validation | mAP: 46.0 -> 47.5 | Helped |
| Add cross-modal GT-Paste to the fused detector (ablation) | PointAugmenting · nuScenes 1/8 validation | mAP: 42.6 -> 47.5 | Helped |
| Add the fade strategy by stopping augmentation in the final 5 epochs (ablation) | PointAugmenting · nuScenes 1/8 validation | mAP: 47.5 -> 48.8 | Helped |
| Add naive GT-Paste to LiDAR-only detection (ablation) | CenterPoint · nuScenes 1/8 validation | mAP: 32.8 -> 37.6 | Helped |
| Replace naive GT-Paste with cross-modal GT-Paste (ablation) | LiDAR-only CenterPoint · nuScenes 1/8 validation | mAP: 37.6 -> 37.4 | No clear effect |
| Reduce input image size to 448 × 800 with late fusion (ablation) | PointAugmenting · nuScenes validation | mAP: 47.5 -> 47.3 | Hurt |
| Use point-wise concatenation instead of late fusion at 448 × 800 (ablation) | PointAugmenting lite · nuScenes validation | mAP: 47.3 -> 46.4 | Hurt |
Closest papers by meaning
- LiDAR-Aug: A General Rendering-Based Augmentation Framework for 3D Object Detection (CVPR 2021 · similarity 0.89)
- VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection (CVPR 2018 · similarity 0.89)
- LiDAR R-CNN: An Efficient and Universal 3D Object Detector (CVPR 2021 · similarity 0.89)
- Pre-training LiDAR-based 3D Object Detectors through Colorization (ICLR 2024 · similarity 0.88)
- Beyond Virtual Points: Depth-Enhanced LiDAR-only 3D Object Detection with Semi-Supervised Learning (Student Abstract) (AAAI 2025 · similarity 0.87)
- STONE: A Submodular Optimization Framework for Active 3D Object Detection (NeurIPS 2024 · similarity 0.87)
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.