A Simple Diffusion Model for 3D Shape(Pointcloud) Generation

This homework implements a simple, unconditional diffusion model for 3D shape generation.

Introduction

We adopt PointTransformer to model pointclouds and predict the noise added. We choose three catogories of 3D shapes: airplanes, chairs and tables for training and generation.

Results

The generated results are shown below:

Generated Airplane and Generation Process

Generated Chair and Generation Process

Generated Table and Generation Process

For more implementation details, please refer to the github page of this project.