Research map / Point clouds / Segmentation and detection
4D Unsupervised Object Discovery
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 4D ClusterNet discovery instead of 10% manual labels | 2D detection on Waymo Open Dataset · Waymo Open Dataset validation | AP^50: 33.8 -> 43.2 | Helped |
| Use ground-truth rather than estimated scene flow | ClusterNet 2D detection · Waymo Open Dataset validation | AP^50: 43.2 -> 51.8 | Helped |
| Use 4D ClusterNet discovery instead of 10% manual labels | ClusterNet 3D instance segmentation · Waymo Open Dataset validation | AP^70: 23.6 -> 26.2 | Helped |
| Add a learned ClusterNet after HDBSCAN motion clustering (ablation) | 3D instance segmentation · Waymo Open Dataset validation | AP^3D: 4.6 -> 6.8 | Helped |
| Add a learned ClusterNet after HDBSCAN motion clustering (ablation) | 2D detection · Waymo Open Dataset validation | AP^2D: 25.1 -> 30.0 | Helped |
| Add 2D-3D cross-modal cues to motion-cue initialization (ablation) | ClusterNet joint optimization · Waymo Open Dataset validation | AP^2D: 30.0 -> 40.4 | Helped |
| Add 2D-3D cross-modal cues to motion-cue initialization (ablation) | ClusterNet joint optimization · Waymo Open Dataset validation | AP^3D: 6.8 -> 25.7 | Helped |
| Add temporal consistency to joint 2D-3D cues (ablation) | ClusterNet joint optimization · Waymo Open Dataset validation | AP^2D: 40.4 -> 43.2 | Helped |
| Add temporal consistency to joint 2D-3D cues (ablation) | ClusterNet joint optimization · Waymo Open Dataset validation | AP^3D: 25.7 -> 26.2 | Helped |
| Exclude anchors with IoU at or below 0.1 from negative samples (ablation) | Initial Faster R-CNN training · Waymo Open Dataset validation | AP^50: 28.2 -> 30.0 | Helped |
| Train the initial localization network longer than 3000 iterations (ablation) | Initial Faster R-CNN training · Waymo Open Dataset validation | AP^50: 30.0 -> 27.9; 30.0 -> 25.3 | Hurt |
| Increase joint optimization from 1 to 3 iterations (ablation) | ClusterNet and localization network · Waymo Open Dataset validation | AP^2D: 30.0 -> 43.2 | Helped |
Closest papers by meaning
- Towards Unsupervised Object Detection From LiDAR Point Clouds (CVPR 2023 · similarity 0.91)
- UNION: Unsupervised 3D Object Detection using Object Appearance-based Pseudo-Classes (NeurIPS 2024 · similarity 0.91)
- Unsupervised Object Detection With LIDAR Clues (CVPR 2021 · similarity 0.91)
- Cross-Modal Distillation for 2D/3D Multi-Object Discovery from 2D Motion (CVPR 2025 · similarity 0.87)
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · similarity 0.87)
- FreePoint: Unsupervised Point Cloud Instance Segmentation (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.