Pointcloud Experiments with ModelNet40. Includes PointNet and GAPNet

The purpose of this repository is to provide a clean implementation of GAPNet. In order to facilitate comparability, it includes PointNet as a baseline. The goal is to come up with a readable implementation of GAPNet that outperforms PointNet.

Always star if you like.

If you enjoy this repo, please give it a star. That would be very appreciated!

Getting in touch.

If you got any bug reports or feature requests, please open an issue here on GitHub.

• Subscribe to my YouTube channel.
• Become a patron.
• Subscribe to my newsletter.
• Visit by homepage/blog.
• Join me on Slack.
• Add me on LinkedIn.
• Add me on Facebook.

Acknowledgements.

Based heavily on:

• https://github.com/TianzhongSong/PointNet-Keras
• https://github.com/FrankCAN/GAPointNet

Those people did some really, really great work!

Networks.

1. PointNet.

Reference: https://arxiv.org/abs/1612.00593

PointNet is included as a baseline.

2. GAPNet.

Reference: https://arxiv.org/abs/1905.08705

GAPNet combines the Graph Neural Network approach with attention.

How to run.

Download ModelNet40 data:

sh download_modelnet40.sh

Prepare data:

python prepare_data.py

Train PointNet:

python train_cls.py pointnet

Train GAPNet draft:

python train_cls.py gapnet_dev

Notes.

This project is under heavy construction.