Research map / Point clouds / Point cloud classification
PointGPT: Auto-regressively Generative Pre-training from Point Clouds
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 generator blocks (0 -> 4) in extractor-generator (ablation) | PointGPT-S · ModelNet40 | accuracy: 93.85 -> 94.21 | Helped |
| Increase generator depth from 4 to 6 (ablation) | PointGPT-S · ModelNet40 | accuracy: 94.21 -> 94.24 | No clear effect |
| Use handcrafted FPFH as generation target instead of coordinates (ablation) | PointGPT-S · ModelNet40 | accuracy: 94.21 -> 94.13 | Hurt |
| Use DGCNN teacher features as target instead of coordinates (ablation) | PointGPT-S · ModelNet40 | accuracy: 94.21 -> 94.35 | Helped |
| Add generation loss as auxiliary objective in fine-tuning (lambda=3) (ablation) | PointGPT-S · ModelNet40 | accuracy: 94.01 -> 94.21 | Helped |
| Raise fine-tuning generation coefficient from 3 to 5 (ablation) | PointGPT-S · ModelNet40 | accuracy: 94.21 -> 94.05 | Hurt |
| Use l1+l2 Chamfer loss instead of l1 only (ablation) | PointGPT-S · ModelNet40 | accuracy: 93.66 -> 94.21 | Helped |
| Add relative direction prompts in generator (ablation) | PointGPT-S · ModelNet40 | accuracy: 93.69 -> 94.21 | Helped |
| Add dual masking (ratio 0.7) to vanilla masking (ablation) | PointGPT-S · ModelNet40 | accuracy: 93.68 -> 94.21 | Helped |
| Raise dual mask ratio from 0.7 to 0.9 (ablation) | PointGPT-S · ModelNet40 | accuracy: 94.21 -> 93.66 | Hurt |
| Scale to PointGPT-L with larger pre-training data and post-pre-training | PointGPT · ScanObjectNN PB_T50_RS | accuracy: 86.9 -> 93.4 | Helped |
| Use PointGPT-S instead of Point-MAE | PointGPT · ModelNet40 few-shot 10-way 10-shot | accuracy: 92.6 -> 92.6 | No clear effect |
Closest papers by meaning
- General Point Model Pretraining with Autoencoding and Autoregressive (CVPR 2024 · similarity 0.91)
- Point Cloud Pre-training with Diffusion Models (CVPR 2024 · similarity 0.88)
- PointRWKV: Efficient RWKV-Like Model for Hierarchical Point Cloud Learning (AAAI 2025 · similarity 0.87)
- PointDGMamba: Domain Generalization of Point Cloud Classification via Generalized State Space Model (AAAI 2025 · similarity 0.86)
- PointMamba: A Simple State Space Model for Point Cloud Analysis (NeurIPS 2024 · similarity 0.86)
- GPAS: Accelerating Convergence of LLM Pretraining via Gradient-Preserving Activation Scaling (NeurIPS 2025 · similarity 0.86)
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.