Research map / Detection and segmentation / 3D detection and driving
Exploring Geometry-Aware Contrast and Clustering Harmonization for Self-Supervised 3D Object Detection
Hanxue Liang, Chenhan Jiang, Dapeng Feng, Xin Chen, Hang Xu, Xiaodan Liang, Wei Zhang, Zhenguo Li, Luc Van Gool
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 GCC-3D pretraining before fine-tuning | CenterPoint-pp · Waymo val (5% labels) | AP/L2: 49.30 -> 52.92 | Improved |
| add GCC-3D pretraining before fine-tuning | CenterPoint-pp · nuScenes val (5% labels) | mAP: 25.79 -> 32.75 | Improved |
| add GCC-3D pretraining before fine-tuning | CenterPoint-voxel · Waymo val (5% labels) | AP/L2: 43.13 -> 44.70 | Improved |
| add GCC-3D pretraining before fine-tuning | CenterPoint-voxel · nuScenes val (100% labels) | mAP: 56.19 -> 57.26 | Improved |
| add two-stage CenterPoint with GCC-3D pretraining | CenterPoint-voxel 2-stage · Waymo val (20% labels) | mAP: 64.63 -> 67.29 | Improved |
| add Geometry-Aware Contrast module alone (ablation) | CenterPoint-pp · nuScenes val (5% labels) | mAP: 25.79 -> 32.56 | Improved |
| add Harmonized Pseudo-Instance Clustering module alone (ablation) | CenterPoint-pp · Waymo val (5% labels) | mAP: 49.30 -> 51.89 | Improved |
| combine geometry-aware contrast with harmonized clustering (ablation) | GCC-3D CenterPoint-pp · nuScenes val (5% labels) | mAP: 32.56 -> 32.75 | Improved |
| combine geometry-aware contrast with harmonized clustering (ablation) | GCC-3D CenterPoint-pp · Waymo val (5% labels) | mAP: 51.89 -> 52.92 | Improved |
| replace hard-label contrast (PointContrast) with geometry-aware soft contrast (ablation) | CenterPoint-pp · nuScenes val (5% labels) | mAP: 30.79 -> 32.56 | Improved |
| add harmonization term to DeepCluster-style pseudo-instance clustering (ablation) | CenterPoint-pp · nuScenes val (5% labels) | mAP: 27.84 -> 30.32 | Improved |
| set harmonization weight alpha to 0.1 instead of 0 (ablation) | CenterPoint-pp · nuScenes val (100% labels) | mAP: 50.48 -> 50.84 | Improved |
Closest papers by meaning
- LIGA-Stereo: Learning LiDAR Geometry Aware Representations for Stereo-Based 3D Detector (ICCV 2021 · similarity 0.89)
- 3D Object Detection with a Self-Supervised Lidar Scene Flow Backbone (ECCV 2022 · similarity 0.89)
- UNION: Unsupervised 3D Object Detection using Object Appearance-based Pseudo-Classes (NeurIPS 2024 · similarity 0.88)
- Seg2Box: 3D Object Detection by Point-Wise Semantics Supervision (AAAI 2025 · similarity 0.88)
- 3DiffTection: 3D Object Detection with Geometry-Aware Diffusion Features (CVPR 2024 · similarity 0.87)
- VSRD: Instance-Aware Volumetric Silhouette Rendering for Weakly Supervised 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.
