Zehaos / mx-pointnet

A MXNet implementation for PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation

mx-pointnet(in progress)

This repository contains a MXNet implementation for the Pointnet. The official implementation is avaliable here.

Note: Only support point cloud classification Currently.

Requirements and Dependencies

• Install MXNet(https://github.com/apache/incubator-mxnet) on a machine with CUDA GPU
• Install CUDNN v5 or v6
• Install h5py for python

Train and Test

• Modify config/cfg.py by your setting
• python train.py
• python test.py

Performance

Classification

Modelnet40	Top1-Acc
Ours	88.0
Reference[1]	89.2

Reference

[1] Charles R. Qi, et al. "PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation" https://arxiv.org/abs/1612.00593 (2016).

[2] Chen, Tianqi, et al. "Mxnet: A flexible and efficient machine learning library for heterogeneous distributed systems." arXiv:1512.01274 (2015).

[3] MXNet training code is based largely on Tusimple's resnet implementation, https://github.com/TuSimple/resnet.mxnet