Research map / Point clouds / Segmentation and detection
Towards Unsupervised Object Detection From LiDAR Point Clouds
Read the paperCheck 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 the proposed detector instead of PP score train | Unsupervised LiDAR object detection · PandaSet, 0–80m all classes | AP @ IoU 0.5: 14.0 -> 29.5 | Helped |
| Use the proposed detector instead of PP score train | Unsupervised LiDAR object detection · Argoverse 2 Sensor, 0–80m all classes | AP @ IoU 0.5: 15.8 -> 24.5 | Helped |
| Initially train on near-range labels (0–40m) instead of full-range labels (0–80m) (ablation) | M1 detector · PandaSet, 0–80m all classes | AP @ IoU 0.5: 12.0 -> 13.0 | Helped |
| Add ray-dropping augmentation (ablation) | Near-range-trained detector · PandaSet, 0–80m all classes | AP @ IoU 0.5: 13.0 -> 14.5 | Helped |
| Add ray-dropping augmentation (ablation) | Near-range-trained detector · PandaSet, 0–80m all classes | AP @ IoU 0.7: 5.3 -> 4.1 | Hurt |
| Add tracking and short-tracklet filtering (ablation) | Ray-drop detector · PandaSet, 0–80m all classes | AP @ IoU 0.5: 14.5 -> 17.8 | Helped |
| Add two rounds of self-training (ablation) | Detector with tracking and short-tracklet filtering · PandaSet, 0–80m all classes | AP @ IoU 0.5: 17.8 -> 26.2 | Helped |
| Add long-tracklet refinement and one further self-training round (ablation) | M5 detector · PandaSet, 0–80m all classes | AP @ IoU 0.5: 26.2 -> 29.5 | Helped |
| Add long-tracklet refinement and one further self-training round (ablation) | M5 detector · PandaSet, 0–80m all classes | AP @ ΔDTC 1.0: 52.3 -> 48.8 | Hurt |
| Use more self-training iterations (ablation) | Proposed detector · PandaSet and Argoverse 2 Sensor | DTC-based recall | Hurt |
Closest papers by meaning
- Unsupervised Object Detection With LIDAR Clues (CVPR 2021 · similarity 0.93)
- 4D Unsupervised Object Discovery (NeurIPS 2022 · similarity 0.91)
- Not All Points Are Equal: Learning Highly Efficient Point-Based Detectors for 3D LiDAR Point Clouds (CVPR 2022 · similarity 0.91)
- Image-to-Lidar Self-Supervised Distillation for Autonomous Driving Data (CVPR 2022 · similarity 0.91)
- UNION: Unsupervised 3D Object Detection using Object Appearance-based Pseudo-Classes (NeurIPS 2024 · similarity 0.89)
- SAFDNet: A Simple and Effective Network for Fully Sparse 3D Object Detection (CVPR 2024 · similarity 0.89)
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.