This repository is an implementation of DGCNN[1] in Chainer.
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
You can simply execute train code with GPU.
python train.py -g 0
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 | ???? |