A-CNN: Annularly Convolutional Neural Networks on Point Clouds
A-CNN
Created by Artem Komarichev, Zichun Zhong, Jing Hua from Department of Computer Science, Wayne State University.
Introduction
Our paper (arXiV) proposes a new approach to define and compute convolution directly on 3D point clouds by the proposed annular convolution.
To appear, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2019
A-CNN usage
We provide the code of A-CNN model that was tested with Tensorflow 1.3.0, CUDA 8.0, and python 3.6 on Ubuntu 16.04. We run all our experiments on a single NVIDIA Titan Xp GPU with 12GB GDDR5X.
-
Classification Task
Download ModelNet-40 dataset first. Point clouds are sampled from meshes with 10K points (XYZ + normals) per shape and provided by PointNet++.