Research map / Point clouds / Segmentation and detection
Unsupervised 3D Point Cloud Representation Learning by Triangle Constrained Contrast for Autonomous Driving
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 |
|---|---|---|---|
| add triplet contrast loss Lc to consistent constraint (ablation) | TriCC · nuScenes 1% fine-tuning, Res16UNet | mIoU: 38.1 -> 39.7 | Helped |
| add cycle shortcut loss Lshortcut (ablation) | TriCC · nuScenes 1% fine-tuning, Res16UNet | mIoU: 39.7 -> 40.8 | Helped |
| use separate projection heads for consistent constraint and contrast instead of shared head (ablation) | TriCC · nuScenes 1% fine-tuning, Res16UNet | mIoU: 39.5 -> 40.8 | Helped |
| add long-term contrast pair (Pt, Pt+2) or (Ct, Pt+2) (ablation) | TriCC · nuScenes 1% fine-tuning, Res16UNet | mIoU: 40.8 -> 40.2 / 39.9 | Hurt |
| add (Pt+1, Pt) contrast pair (ablation) | TriCC · nuScenes 1% fine-tuning, Res16UNet | mIoU: 40.6 -> 40.8 | Helped |
| pretrain TriCC for 20 epochs instead of baselines' 50 epochs | TriCC · nuScenes 1% fine-tuning, Res16UNet | mIoU: 40.8 vs SLidR 38.2 (50 epochs) | Helped |
| TriCC pretraining (triangle constraint + triplet contrast) | Res16UNet · nuScenes linear probing | mIoU: 37.8 vs SLidR 38.0 | No clear effect |
| TriCC pretraining | Res16UNet · nuScenes fine-tuning 100% | mIoU: +1.0 over SLidR (75.6 vs 74.6) | Helped |
| TriCC pretraining | VoxelNet · nuScenes fine-tuning 1% | mIoU: 34.0 vs SLidR 32.1 | Helped |
| TriCC pretraining | Res16UNet · SemanticKITTI fine-tuning 1% | mIoU: 45.9 vs SLidR 44.6 | Helped |
| TriCC pretraining | VoxelNet + CenterPoint · nuScenes 5% detection | mAP: 44.6 vs SLidR 43.3 | Helped |
| TriCC pretraining | SECOND and PV-RCNN · KITTI 100% detection | mAP: +2.5 (SECOND), +2.7 (PV-RCNN) mAP over random | Helped |
Closest papers by meaning
- Temporal Consistent 3D LiDAR Representation Learning for Semantic Perception in Autonomous Driving (CVPR 2023 · similarity 0.91)
- VLM2Scene: Self-Supervised Image-Text-LiDAR Learning with Foundation Models for Autonomous Driving Scene Understanding (AAAI 2024 · similarity 0.91)
- TREND: Unsupervised 3D Representation Learning via Temporal Forecasting for LiDAR Perception (NeurIPS 2025 · similarity 0.89)
- HiLoTs: High-Low Temporal Sensitive Representation Learning for Semi-Supervised LiDAR Segmentation in Autonomous Driving (CVPR 2025 · similarity 0.88)
- UniPAD: A Universal Pre-training Paradigm for Autonomous Driving (CVPR 2024 · similarity 0.88)
- Visual Point Cloud Forecasting enables Scalable Autonomous Driving (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.