https://arxiv.org/abs/1807.03039
後で実装解説を書きます。
- Python 3
- Ubuntu
- Chainer 4
- Extremely huge amount of GPU memory
- python-tabulate
- scipy
pip3 install chainer cupy h5py
pip3 install tabulate
Large-scale CelebFaces Attributes (CelebA) Dataset
cd run
python3 train.py -dataset /home/user/dataset/celeba-64x64-images-npy/ -b 4 -depth 32 -levels 4 -nn 512 -bits 5 -ext npy
cd run/experiments
python3 change_channels.py -snapshot-1 ../snapshot_512 -snapshot-2 ../snapshot_256 -snapshot-3 ../snapshot_128
cd run/experiments
python3 change_temperature.py -snapshot ../snapshot
cd run/experiments
python3 interpolation.py -snapshot ../snapshot -dataset /home/user/dataset/celeba-64x64-images-npy/ -ext npy -temp 1
cd run/experiments
python3 generate.py -snapshot ../snapshot_hq --temperature 0.7
8 GPUs with 16GB memory were used in this experiment and it took 4 days to get the result.
cd run/experiments
python3 random_walk.py -snapshot ../snapshot_hq --temperature 0.7
