Research map / Point clouds / Point cloud classification
Point-BERT: Pre-Training 3D Point Cloud Transformers With Masked Point Modeling
Check 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 |
|---|---|---|---|
| pre-train a standard Transformer with masked point modeling (ablation) | Transformer · ModelNet40 | accuracy (%): 91.41 -> 92.58 | Helped |
| add Point Patch Mixing to MPM pre-training (ablation) | Transformer · ModelNet40 | accuracy (%): 92.58 -> 92.91 | Helped |
| add MoCo contrastive learning to MPM and Point Patch Mixing (ablation) | Transformer · ModelNet40 | accuracy (%): 92.91 -> 93.24 | Helped |
| add Point-BERT pre-training | standard Transformer · ModelNet40 | accuracy (%): 91.4 -> 93.2 | Helped |
| increase Point-BERT input points from 1k to 4k | Point-BERT · ModelNet40 | accuracy (%): 93.2 -> 93.4 | Helped |
| increase Point-BERT input points from 4k to 8k | Point-BERT · ModelNet40 | accuracy (%): 93.4 -> 93.8 | Helped |
| use Point-BERT pre-training instead of training from scratch | Transformer · ScanObjectNN OBJ-BG, OBJ-ONLY, PB-T50-RS | accuracy (%): 79.86 -> 87.43; 80.55 -> 88.12; 77.24 -> 83.07 | Helped |
| use Point-BERT representations instead of Transformer-OcCo | Transformer · ModelNet40 few-shot classification | accuracy (%): 94.0 ± 3.6 -> 94.6 ± 3.1; 95.9 ± 2.3 -> 96.3 ± 2.7; 89.4 ± 5.1 -> 91.0 ± 5.4; 92.4 ± 4.6 -> 92.7 ± 5.1 | Helped |
| use Point-BERT pre-training instead of vanilla Transformer training | Transformer · ShapeNetPart | mIoU_C (%): 83.42 -> 84.11 | Helped |
| use Point-BERT pre-training instead of vanilla Transformer training | Transformer · ShapeNetPart | mIoU_I (%): 85.1 -> 85.6 | Helped |
| replace block masking with random masking (ablation) | Model B (MPM) · ModelNet40 | accuracy (%): 92.58 -> 92.34 | Hurt |
| increase the mask ratio from [0.25, 0.45] to [0.55, 0.85] (ablation) | Model B (MPM) · ModelNet40 | accuracy (%): 92.58 -> 92.52 | Hurt |
Closest papers by meaning
- General Point Model Pretraining with Autoencoding and Autoregressive (CVPR 2024 · similarity 0.90)
- Frozen CLIP Transformer Is an Efficient Point Cloud Encoder (AAAI 2024 · similarity 0.89)
- 3D Feature Prediction for Masked-AutoEncoder-Based Point Cloud Pretraining (ICLR 2024 · similarity 0.89)
- Towards Compact 3D Representations via Point Feature Enhancement Masked Autoencoders (AAAI 2024 · similarity 0.89)
- Point Transformer V3: Simpler Faster Stronger (CVPR 2024 · similarity 0.88)
- Point Cloud Pre-training with Diffusion Models (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.