Multi-Scale VQVAE for Point Cloud Processing

This project implements a Vector Quantized Variational Autoencoder (VQVAE) for processing point clouds at multiple scales. The multi-scale approach allows the model to capture both fine-grained details and broader structural information from point cloud data.

Key Features

• Multi-scale point cloud representation
• VQVAE architecture for efficient encoding and decoding

Dataset

The point clouds are uniformly sampled from meshes from ShapeNetCore dataset (version 2) and use the official split. Please use this link to download the ShapeNet point cloud. The point cloud should be placed into data directory.

File Structure

• vqvae_model.py: Contains the VQVAE model implementation, including Encoder, Decoder, and VectorQuantizer classes.
• data_processing.py: Handles data loading and preprocessing for the ShapeNetV2 dataset.
• train.py: Implements the training loop and hyperparameter settings.
• README.md: Project documentation (this file).