Research map / Point clouds / Point cloud classification
Masked Autoencoders for Point Cloud Self-Supervised Learning
Yatian Pang, Wenxiao Wang, Francis E.H. Tay, Wei Liu, Yonghong Tian, Li Yuan
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 Point-MAE pretraining vs training from scratch | Transformer · ScanObjectNN PB-T50-RS | accuracy: 77.24 -> 85.18 | Improved |
| add Point-MAE pretraining vs training from scratch | Transformer · ScanObjectNN OBJ-BG | accuracy: 79.86 -> 90.02 | Improved |
| add Point-MAE pretraining vs training from scratch | Transformer · ModelNet40 | accuracy: 91.4% -> 93.8% | Improved |
| increase input points from 1024 to 8192 | Point-MAE · ModelNet40 | accuracy: 93.8% -> 94.04% | Improved |
| replace Point-BERT pretraining with Point-MAE | Transformer · ScanObjectNN PB-T50-RS | accuracy: 83.07 -> 85.18 | Improved |
| replace Point-BERT pretraining with Point-MAE | Transformer · ModelNet40 | accuracy: 93.2% -> 93.8% | Improved |
| replace Point-BERT with Point-MAE for few-shot | Transformer · ModelNet40 few-shot | accuracy: 92.7 -> 95.0 | Improved |
| add Point-MAE pretraining on part segmentation | Transformer · ShapeNetPart | mIoU_I: 85.1 -> 86.1 | Improved |
| use random masking instead of block masking | Point-MAE · ModelNet40 | accuracy: 92.67 -> 93.19 | Improved |
| raise random masking ratio from 40% to 60% (ablation) | Point-MAE · ModelNet40 | accuracy: 92.46 -> 93.19 | Improved |
| raise random masking ratio from 60% to 90% (ablation) | Point-MAE · ModelNet40 | accuracy: 93.19 -> 92.63 | Hurt |
| shift mask tokens to the lightweight decoder (ablation) | Point-MAE · ModelNet40 | accuracy: 92.14% -> 93.19% | Improved |
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 0.91)
- RI-MAE: Rotation-Invariant Masked AutoEncoders for Self-Supervised Point Cloud Representation Learning (AAAI 2025 · similarity 0.91)
- Towards Compact 3D Representations via Point Feature Enhancement Masked Autoencoders (AAAI 2024 · similarity 0.90)
- Self-Guided Masked Autoencoder (NeurIPS 2024 · similarity 0.89)
- PCP-MAE: Learning to Predict Centers for Point Masked Autoencoders (NeurIPS 2024 · similarity 0.89)
- General Point Model Pretraining with Autoencoding and Autoregressive (CVPR 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.
