PCT: Point Cloud Transformer

pct is a simple PyTorch-based package for learning latent representations from 3D point clouds.
It includes a minimal Point Cloud Transformer (PCT) model, dataset utilities, and training scripts.
The latent embeddings can be used for regression or other downstream tasks.

---

Features

• Dummy point cloud dataset generator (noisy spheres with variable radius).
• Collate functions for batching point clouds.
• Transformer-based point cloud encoder with regression head.
• Training utilities with PyTorch DataLoader.
• 3D visualization of point clouds using Plotly.

---

Use Cases:

• Example 1: Given a 3D point cloud from an object, classify the object into a lifing category, e.g., 'healthy', 'degraded', 'failure'.
  • In aerospace applications, this may enable point clouds obtained from scans to be used to train a model that categorises components such that they can be batched appropriately, or maintained accordingly.
• Example 2: Given a point cloud representation of a component, classify this component as geometrically suitable for a manufacturing method such as 'TRUE' or 'FALSE'.

Installation

Clone the repo and install locally:

git clone https://github.com/your-username/pct.git
cd pct
pip install .

Citations

https://arxiv.org/pdf/2012.09688