Research map / Detection and segmentation / 3D detection and driving
Unsupervised Object Detection With LIDAR Clues
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 LiDAR point-cloud features for segment labeling (ablation) | Unsupervised object detection · Waymo Open validation | AP: 26.3 vs 26.9; 18.7 vs 21.1; 3.3 vs 5.2 | Helped |
| Generate candidate segments from 3D point clouds rather than 2D MCG proposals (ablation) | Unsupervised object detection · Waymo Open validation | AP: 11.0 vs 26.9; 0.0 vs 21.1; 0.0 vs 5.2 | Helped |
| Use simplified Equalization Loss instead of softmax loss (ablation) | Segment labeling and detector classification · Waymo Open validation | AP: 25.4 -> 28.7; 20.3 -> 28.5; 2.0 -> 1.0 | Helped |
| Use simplified Equalization Loss instead of sigmoid loss (ablation) | Segment labeling and detector classification · Waymo Open validation | AP: 23.8 -> 28.7; 20.7 -> 28.5; 0.6 -> 1.0 | Mixed |
| Increase iterative labeling rounds from 1 to 10 (ablation) | Unsupervised object detection · Waymo Open validation | AP: vehicles 26.9 -> 28.7; pedestrians 21.1 -> 28.5; cyclists 5.2 -> 1.0 | Mixed |
| Use ImageNet supervised rather than MoCo v2 pretraining (ablation) | Manual-annotation detector with 15k images · Waymo Open validation | AP: vehicles 29.2 -> 30.1; pedestrians 22.7 -> 22.2; cyclists 3.9 -> 4.6 | Mixed |
| Increase training data from 10% to 100% of video sequences (ablation) | Unsupervised object detection · Waymo Open validation | AP: vehicles 23.1 -> 26.9; pedestrians 15.5 -> 21.1; cyclists 1.6 -> 5.2 | Helped |
| Use Waymo unsupervised detector pretraining rather than random initialization | ResNet-50 FPN transfer model · Cityscapes | AP bbox: 30.6 -> 36.1 | Helped |
| Use Waymo unsupervised detector pretraining rather than ImageNet classification pretraining | ResNet-50 FPN transfer model · Cityscapes | AP bbox: 37.5 -> 36.1 | Hurt |
| Use Waymo unsupervised detector pretraining rather than random initialization | ResNet-50 FPN transfer model · PASCAL VOC | AP bbox: 45.3 -> 53.8 | Helped |
| Use Waymo unsupervised detector pretraining rather than ImageNet classification pretraining | ResNet-50 FPN transfer model · PASCAL VOC | AP bbox: 54.6 -> 53.8 | Hurt |
| Vary the foreground threshold across 0.80–0.999 (ablation) | Iterative segment labeling · Waymo Open validation | AP: vehicles 23.0–27.7; pedestrians 11.5–22.6; cyclists 0.0–5.9 | Mixed |
Closest papers by meaning
- UNION: Unsupervised 3D Object Detection using Object Appearance-based Pseudo-Classes (NeurIPS 2024 · similarity 0.90)
- Learning to Detect Objects from Multi-Agent LiDAR Scans without Manual Labels (CVPR 2025 · similarity 0.88)
- MixSup: Mixed-grained Supervision for Label-efficient LiDAR-based 3D Object Detection (ICLR 2024 · similarity 0.88)
- Unsupervised Object Detection with Theoretical Guarantees (NeurIPS 2024 · similarity 0.88)
- SAFDNet: A Simple and Effective Network for Fully Sparse 3D Object Detection (CVPR 2024 · similarity 0.87)
- Commonsense Prototype for Outdoor Unsupervised 3D Object Detection (CVPR 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.