This repo contains code for creating a point cloud that, when viewed from different angles, forms different images. It works by initializing a random point cloud and optimizing with PyTorch3D using a simple image difference loss function.
| Input 1 | Input 2 | Input 2 |
|---|---|---|
 |
 |
 |
| Point Cloud Perspective 1 | Point Cloud Perspective 2 | Point Cloud Perspective 3 |
 |
 |
 |
PerspectivePoints.mp4
PyTorch3D requires conda. Install the environment pytorch3d_notebooks.yml.