Research map / Point clouds / Segmentation and detection
Self-Supervised Pretraining of 3D Features on Any Point-Cloud
Zaiwei Zhang, Rohit Girdhar, Armand Joulin, Ishan Misra
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 |
|---|---|---|---|
| self-supervised DepthContrast pretraining instead of scratch | VoteNet PointNet++ · S3DIS detection | AP25: 31.2 -> 43.3 | Improved |
| self-supervised DepthContrast pretraining instead of scratch | VoteNet PointNet++ · ScanNet detection | AP25: 58.6 -> 61.3 | Improved |
| self-supervised DepthContrast pretraining instead of scratch | VoteNet PointNet++ · SUNRGBD detection | AP25: 57.4 -> 60.4 | Improved |
| DepthContrast instead of supervised ScanNet detection pretraining | VoteNet PointNet++ · SUNRGBD detection | AP25: 59.1 -> 60.4 | Improved |
| DepthContrast instead of supervised ScanNet detection pretraining | VoteNet PointNet++ · S3DIS detection | AP25: 48.5 -> 43.3 | Hurt |
| add across-format loss to within-format only (ablation) | joint point and voxel pretraining · S3DIS voxel transfer | transfer score: 66.5 -> 70.6 | Improved |
| add across-format loss to within-format only (ablation) | joint point and voxel pretraining · ScanNet point transfer | AP25: 61.3 -> 62.2 | Improved |
| add within-format loss to across-format only (ablation) | joint point and voxel pretraining · SUNRGBD point transfer | AP25: 60.0 -> 60.7 | Improved |
| pretrain on combined Redwood-vid and ScanNet-vid and use larger model with H3DNet | H3DNet · ScanNet detection | AP25: 67.2 -> 69.0 | Improved |
| pretrain on combined Redwood-vid and ScanNet-vid and use larger model with H3DNet | H3DNet · SUNRGBD detection | AP25: 60.1 -> 63.5 | Improved |
| DepthContrast pretraining for VoteNet instead of PointContrast | VoteNet · ScanNet detection | AP25: 59.2 -> 64.0 | Improved |
| increase model width with scratch training (ablation) | PointNet++ · S3DIS detection | AP25 | No clear effect |
Closest papers by meaning
- Self-Supervised Pretraining for Large-Scale Point Clouds (NeurIPS 2022 · similarity 0.95)
- 3D Object Detection with a Self-Supervised Lidar Scene Flow Backbone (ECCV 2022 · similarity 0.89)
- A Unified Approach to Interpreting Self-supervised Pre-training Methods for 3D Point Clouds via Interactions (CVPR 2025 · similarity 0.88)
- Occlusion-aware Text-Image-Point Cloud Pretraining for Open-World 3D Object Recognition (CVPR 2025 · similarity 0.88)
- Point Cloud Pre-training with Diffusion Models (CVPR 2024 · similarity 0.87)
- Multi-View Representation is What You Need for Point-Cloud Pre-Training (ICLR 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.
