Dynamic Graph CNN using chainer

Introduce

This repository is an implementation of DGCNN[1] in Chainer.

Installation

Please install Chainer (and cupy if you want to use GPU) beforehand.
Furthermore, version that I tested operation is described on comments.

# chainer version 5.3.0
pip install chainer
# cupy-cuda100 version 5.4.0
pip install cupy-cuda100

Also, some extension library is used in some of the code,

# Chainer Chemistry version 0.5.0
git clone https://github.com/pfnet-research/chainer-chemistry.git
pip install -e chainer-chemistry
# ChainerEX version 0.0.1
git clone https://github.com/corochann/chainerex.git
pip install -e chainerex

Train

You can simply execute train code with GPU.

python train.py -g 0

Result

2019/04/10(YYYY/MM/DD), this implementation is incomplete. Chainer=our implementation, TensorFlow=author implementation.

framework	main/loss	main/accuracy	validation/main/loss	validation/main/accuracy	elapsed_time
Chainer	0.0147	0.9948	0.7025	0.8906	116486.35(s) ≒ 32.36(h)
TensorFlow	???	0.9741	????	0.9111	????

Reference

1. Yue Wang, Yongbin Sun, Ziwei Liu, Sanjay E. Sarma, Michael M. Bronstein, Justin M. Solomon. Dynamic Graph CNN for Learning on Point Clouds. 2018.
2. WangYueFt. dgcnn. (access:2019/03/31)
3. corochann. chainer-pointnet. (access:2019/03/31)