Research map / Point clouds / Point cloud classification
PointContrast: Unsupervised Pre-training for 3D Point Cloud Understanding
Saining Xie, Jiatao Gu, Demi Guo, Charles R. Qi, Leonidas Guibas, Or Litany
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 PointContrast unsupervised pre-training on ScanNet pairs | SR-UNet · S3DIS Area 5 segmentation | mIoU: 68.2 -> 70.9 | Improved |
| add PointContrast pre-training | SR-UNet · ScanNet segmentation | mIoU: 72.2 -> 74.1 | Improved |
| add PointContrast pre-training | VoteNet with SR-UNet · ScanNet detection | mAP@0.5: 35.4 -> 38.0 | Improved |
| add PointContrast pre-training | VoteNet with SR-UNet · SUN RGB-D detection | mAP@0.5: 31.7 -> 34.8 | Improved |
| add PointContrast pre-training | SR-UNet · Synthia4D segmentation | mIoU: 79.8 -> 83.1 | Improved |
| add PointContrast pre-training with 1% labeled data | SR-UNet · ShapeNet classification | accuracy: 62.2 -> 66.2 | Improved |
| add PointContrast pre-training with all training data | SR-UNet · ShapeNet classification | accuracy: 85.1 -> 85.7 | Improved |
| add PointContrast pre-training with 1% labeled data | SR-UNet · ShapeNet part segmentation | IoU: 71.8 -> 74.0 | Improved |
| add ShapeNet supervised pre-training before fine-tuning | SR-UNet · S3DIS segmentation | training curve | Hurt |
| use PointInfoNCE instead of hardest-contrastive loss (ablation) | PointContrast · Synthia4D segmentation | mIoU: 82.6 -> 83.1 | Improved |
| use PointInfoNCE instead of hardest-contrastive loss (ablation) | PointContrast · S3DIS Area 5 segmentation | mIoU: 70.9 -> 70.3 | Hurt |
| use multi-view partial scans instead of whole scene as a single view (ablation) | PointContrast · S3DIS Area 5 segmentation | mIoU: 68.17 -> 68.35 | Improved |
Closest papers by meaning
- Multi-View Representation is What You Need for Point-Cloud Pre-Training (ICLR 2024 · similarity 0.91)
- Sonata: Self-Supervised Learning of Reliable Point Representations (CVPR 2025 · similarity 0.89)
- Point Cloud Pre-training with Diffusion Models (CVPR 2024 · similarity 0.89)
- Exploring Vision Semantic Prompt for Efficient Point Cloud Understanding (ICML 2025 · similarity 0.89)
- LinNet: Linear Network for Efficient Point Cloud Representation Learning (NeurIPS 2024 · similarity 0.89)
- UniPre3D: Unified Pre-training of 3D Point Cloud Models with Cross-Modal Gaussian Splatting (CVPR 2025 · 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.
