Research map / Point clouds / Point cloud classification
Spatio-Temporal Self-Supervised Representation Learning for 3D Point Clouds
Siyuan Huang, Yichen Xie, Song-Chun Zhu, Yixin Zhu
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 rotation transformation (ablation) | STRL + PointNet · ModelNet40 linear (ShapeNet pretrain, PointNet) | accuracy: 87.8% -> 88.3% | Improved |
| add scaling transformation (ablation) | STRL + PointNet · ModelNet40 linear (ShapeNet pretrain, PointNet) | accuracy: 87.9% -> 88.3% | Improved |
| add translation transformation (ablation) | STRL + PointNet · ModelNet40 linear (ShapeNet pretrain, PointNet) | accuracy: 87.2% -> 88.3% | Improved |
| add all synthetic view transformations (ablation) | STRL + PointNet · ModelNet40 linear (ShapeNet pretrain, PointNet) | accuracy: 85.5% -> 88.3% | Improved |
| add natural-sequence sampling of input pairs (ablation) | STRL + PV-RCNN (frozen backbone) · KITTI Car (Easy) | mAP40: 90.17 -> 91.08 | Improved |
| add natural-sequence sampling of input pairs (ablation) | STRL + PV-RCNN (frozen backbone) · KITTI Car (Moderate) | mAP40: 81.21 -> 81.63 | Improved |
| add random cutout (ablation) | STRL + PointNet · ModelNet40 linear (ShapeNet pretrain, PointNet) | accuracy: 87.4% -> 87.5% | Improved |
| add random crop (ablation) | STRL + PointNet · ModelNet40 linear (ShapeNet pretrain, PointNet) | accuracy: 86.1% -> 87.5% | Improved |
| add random cutout on top of crop (ablation) | STRL + PointNet · ModelNet40 linear (ShapeNet pretrain, PointNet) | accuracy: 88.1% -> 88.3% | Improved |
| pretrain on ShapeNet instead of ScanNet (ablation) | STRL + DGCNN · ModelNet40 linear | accuracy: 90.4% -> 90.9% | Improved |
| pretrain on ShapeNet instead of ScanNet (ablation) | STRL + VoteNet · SUN RGB-D | mAP@0.25: 58.2 -> 59.2 | Improved |
| freeze pretrained backbone during fine-tuning | STRL + PV-RCNN · KITTI Pedestrian 3D | mAP40: 57.80 -> 39.62 | Hurt |
Closest papers by meaning
- Sonata: Self-Supervised Learning of Reliable Point Representations (CVPR 2025 · similarity 0.90)
- Masked Scene Modeling: Narrowing the Gap Between Supervised and Self-Supervised Learning in 3D Scene Understanding (CVPR 2025 · similarity 0.88)
- PointGLR: Unsupervised Structural Representation Learning of 3D Point Clouds (TPAMI 2022 · similarity 0.88)
- Self-Supervised Pretraining for Large-Scale Point Clouds (NeurIPS 2022 · similarity 0.88)
- Spatiotemporal Self-Supervised Learning for Point Clouds in the Wild (CVPR 2023 · similarity 0.88)
- PSA-SSL: Pose and Size-aware Self-Supervised Learning on LiDAR Point Clouds (CVPR 2025 · 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.
