Research map / Point clouds / Point cloud classification
Point-M2AE: Multi-scale Masked Autoencoders for Hierarchical Point Cloud Pre-training
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 hierarchical pre-training (vs scratch) (ablation) | Point-M2AE · ModelNet40 | accuracy: 92.5 -> 94.0 | Helped |
| add hierarchical pre-training (vs scratch) (ablation) | Point-M2AE · ScanObjectNN PB-T50-RS | accuracy: 83.9 -> 86.4 | Helped |
| add hierarchical pre-training (vs scratch) (ablation) | Point-M2AE · ModelNet40 10-way 20-shot | accuracy: 91.2 -> 95.0 | Helped |
| add hierarchical pre-training (vs scratch) (ablation) | Point-M2AE · ShapeNetPart | accuracy: 85.4 -> 86.5 | Helped |
| add multi-scale consistent masking (vs independent random masks per scale) (ablation) | Point-M2AE · ModelNet40 linear SVM | accuracy: -4.5% without | Helped |
| add hierarchical encoder and decoder with skip connections (ablation) | Point-M2AE · ModelNet40 linear SVM | accuracy | Helped |
| add local spatial attention (ablation) | Point-M2AE fine-tuning · ScanObjectNN PB-T50-RS | accuracy: 85.82 -> 86.43 | Helped |
| use max + average pooling instead of class token (ablation) | Point-M2AE fine-tuning · ModelNet40 | accuracy: 93.4 -> 94.0 | Helped |
| use L2-norm Chamfer loss instead of EMD (ablation) | Point-M2AE pre-training · ModelNet40 linear SVM | accuracy: 91.9 -> 92.9 | Helped |
| add EMD to L2-norm Chamfer loss (ablation) | Point-M2AE pre-training · ModelNet40 linear SVM | accuracy: 92.9 -> 92.4 | Hurt |
| use 3-stage decoder instead of 2-stage (ablation) | Point-M2AE · ModelNet40 linear SVM | accuracy | Hurt |
| multi-scale MAE vs single-scale Point-BERT | pretrained point transformer · ScanObjectNN OBJ-BG | accuracy: 87.43 -> 91.22 | Helped |
Closest papers by meaning
- 3D Feature Prediction for Masked-AutoEncoder-Based Point Cloud Pretraining (ICLR 2024 · similarity 0.93)
- Towards Compact 3D Representations via Point Feature Enhancement Masked Autoencoders (AAAI 2024 · similarity 0.93)
- RI-MAE: Rotation-Invariant Masked AutoEncoders for Self-Supervised Point Cloud Representation Learning (AAAI 2025 · similarity 0.91)
- Multi-Scale Neighborhood Occupancy Masked Autoencoder for Self-Supervised Learning in LiDAR Point Clouds (CVPR 2025 · similarity 0.90)
- Adaptive Siamese Masked Autoencoder with Global Optimization for Unsupervised Point Cloud Shape Correspondence (AAAI 2025 · similarity 0.90)
- PCP-MAE: Learning to Predict Centers for Point Masked Autoencoders (NeurIPS 2024 · similarity 0.90)
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.