winterchaufr/Open3D-PointNet

Open3D PointNet implementation with PyTorch

Open3D-PointNet

PointNet implementation and visualization with Open3D, an open-source library that supports rapid development of software that deals with 3D data. As part of the Open3D ecosystem, this repository demonstrates how Open3D can be used for ML/DL research projects.

This repository is forked from fxia22's PyTorch implementation.

Changelog

1. Added CPU support for non-cuda-enabled devices.
2. Used Open3D point cloud loader for loading PointNet datasets (datasets.py).
3. Added example for PointNet inference with Open3D Jupyter visualization (open3d_pointnet_inference.ipynb).
4. Added example for native OpenGL visualization with Open3D (open3d_visualize.py).

Setup

# Install Open3D, must be v0.4.0 or above for Jupyter support
pip install open3d-python

# Install PyTorch
# Follow: https://pytorch.org/

# Install other dependencies
pip install -r requirements.txt

Now, launch

jupyter notebook

and run open3d_pointnet_inference.ipynb. All datasets and pre-trained models shall be downloaded automatically. If you run into issues downloading files, please run download.py separately.