Research map / Point clouds / Segmentation and detection
Implicit Surface Contrastive Clustering for LiDAR Point Clouds
Zaiwei Zhang, Min Bai, Erran Li
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 |
|---|---|---|---|
| pretrain with ISCC instead of SSPL | 3D sparse U-Net · SemanticKITTI, 1% fine-tuning | mIoU: 42.5 -> 45.1 | Improved |
| pretrain with ISCC instead of SSPL | 3D sparse U-Net · Waymo Open Dataset, 1% fine-tuning | mIoU: 44.8 -> 46.0 | Improved |
| pretrain with ISCC instead of DepthContrast | Part-A2 3D detector, KITTI Car Moderate · KITTI, 5% fine-tuning | mAP R11: 65.0 -> 68.9 | Improved |
| add occupancy prediction with 8 neighboring voxel features (ablation) | occupancy-only pretraining · SemanticKITTI, 1% fine-tuning | mIoU: 38.9 -> 40.5 | Improved |
| increase occupancy aggregation from 8 to 16 neighboring voxel features (ablation) | occupancy-only pretraining · SemanticKITTI, 1% fine-tuning | mIoU: 40.5 -> 43.0 | Improved |
| increase occupancy aggregation from 16 to 32 neighboring voxel features (ablation) | occupancy-only pretraining · SemanticKITTI, 1% fine-tuning | mIoU: 43.0 -> 43.6 | Improved |
| add multi-view frames with 2-frame separation instead of single-view input (ablation) | ISCC · SemanticKITTI, 1% fine-tuning | mIoU: 43.7 -> 44.0 | Improved |
| increase multi-view frame separation from 2 to 5 (ablation) | ISCC · SemanticKITTI, 1% fine-tuning | mIoU: 44.0 -> 45.1 | Improved |
| add the global clustering loss to local-only contrastive learning (ablation) | ISCC · SemanticKITTI, 1% fine-tuning | mIoU: 41.3 -> 43.6 | Improved |
| add the local contrastive loss to global-only clustering (ablation) | ISCC · SemanticKITTI, 1% fine-tuning | mIoU: 41.7 -> 43.6 | Improved |
| increase the maximum cluster count from 100 to 200 (ablation) | full contrastive clustering · SemanticKITTI, 1% fine-tuning | mIoU: 43.6 -> 43.5 | No clear effect |
Closest papers by meaning
- Spatiotemporal Self-Supervised Learning for Point Clouds in the Wild (CVPR 2023 · similarity 0.92)
- Self-Supervised Pretraining for Large-Scale Point Clouds (NeurIPS 2022 · similarity 0.92)
- PSA-SSL: Pose and Size-aware Self-Supervised Learning on LiDAR Point Clouds (CVPR 2025 · similarity 0.89)
- Fine-grained Image-to-LiDAR Contrastive Distillation with Visual Foundation Models (NeurIPS 2024 · similarity 0.88)
- TREND: Unsupervised 3D Representation Learning via Temporal Forecasting for LiDAR Perception (NeurIPS 2025 · similarity 0.88)
- VLM2Scene: Self-Supervised Image-Text-LiDAR Learning with Foundation Models for Autonomous Driving Scene Understanding (AAAI 2024 · similarity 0.88)
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.
