This code is the tensorflow implementation of our preprinted paper, RGCNN: Regularized Graph CNN for Point Cloud Segmentation, ACM MultiMedia, 2018. Camera-ready version will be updated soon.

This code runs on tensorflow 1.4 and python 3.6 with additional library such as h5py. We borrow the framework of cnn_graph.

It requires original ModelNet40 and ShapeNet data, which can be downloaded here for segmentation and here for classification. You can use the tool provided by pointnet to convert the data to numpy array. We also provide our processed one but we don't guarantee its compatibility.

The train.py is quite easy to read, I'm sure you can run and test it smoothly.

We test our code on jupyter notebook at first, so I suspect some part could be missing. If something went wrong, please contact me by email.