Research map / Point clouds / Point cloud classification
Masked Discrimination for Self-Supervised Learning on Point Clouds
Haotian Liu, Mu Cai, Yong Jae Lee
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 MaskPoint pretraining vs scratch | PointViT · ModelNet40 | OA: 91.4 -> 93.8 | Improved |
| add MaskPoint pretraining vs scratch | PointViT · ScanObjectNN PB | OA: 77.2 -> 84.6 | Improved |
| replace Point-BERT pretraining with MaskPoint | PointViT · ScanObjectNN PB | OA: 83.1 -> 84.6 | Improved |
| replace OcCo pretraining with MaskPoint | PointViT · ModelNet40 | OA: 92.1 -> 93.8 | Improved |
| add MaskPoint pretraining on part segmentation | PointViT · ShapeNetPart | instance mIoU: 85.1 -> 86.0 | Improved |
| add MaskPoint pretraining to 3DETR | 3DETR · ScanNet detection | AP25 / AP50: +1.3AP25 and +2.7AP50 | Improved |
| add Point-BERT pretraining to 3DETR | 3DETR · ScanNet detection | AP25: 62.1 -> 61.0 | No clear effect |
| increase 3DETR encoder layers from 3 to 12 with MaskPoint | 3DETR · ScanNet detection | AP25 / AP50: +2.1AP25 and +4.2AP50 | Improved |
| raise mask ratio from 0.25 to 0.90 (random) (ablation) | MaskPoint · ScanObjectNN PB | OA: 83.2 -> 84.6 | Improved |
| use random masking instead of block masking at 0.90 (ablation) | MaskPoint · ScanObjectNN PB | OA: 84.1 -> 84.6 | Improved |
| use 256 decoder queries instead of 64 (ablation) | MaskPoint · ScanObjectNN PB | OA: 83.7 -> 84.6 | Improved |
| increase decoder queries from 256 to 1024 (ablation) | MaskPoint · ScanObjectNN PB | OA: 84.6 -> 83.9 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Point-BERT — Point-BERT: Pre-Training 3D Point Cloud Transformers With Masked Point Modeling (CVPR 2022) · better on 7 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- 3D Feature Prediction for Masked-AutoEncoder-Based Point Cloud Pretraining (ICLR 2024 · similarity 0.92)
- RI-MAE: Rotation-Invariant Masked AutoEncoders for Self-Supervised Point Cloud Representation Learning (AAAI 2025 · similarity 0.89)
- Towards Compact 3D Representations via Point Feature Enhancement Masked Autoencoders (AAAI 2024 · similarity 0.89)
- Multi-Scale Neighborhood Occupancy Masked Autoencoder for Self-Supervised Learning in LiDAR Point Clouds (CVPR 2025 · similarity 0.89)
- PCP-MAE: Learning to Predict Centers for Point Masked Autoencoders (NeurIPS 2024 · similarity 0.88)
- Adaptive Siamese Masked Autoencoder with Global Optimization for Unsupervised Point Cloud Shape Correspondence (AAAI 2025 · 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.
