Conditional DRAGAN(cGAN & WGAN based) code(& model) repo.
- Generator 4-layer-512-ReluMLP(last layer is tanh)
- Discriminator 4-layer-512-ReluMLP.
NOTE: Not include any normalization.
Question is always welcome, Pls add some issue.
Its cGAN. Concat Noise z and Label c(onehot), then linear layer(Fully-Coneccted Layer) handle it.
python cdragan -fmnist --gpu 0 --epoch 100 --out result
clone repo and run as follows:
nvidia-docker build -t rizaudo/cdragan-vis .
nvidia-docker run -it -v (pwd):/dic rizaudo/cdragan-vis /usr/bin/python3 /dic/visualizer.py -o /dic/visu --gendir /dic/models/gen_iter_18000.npz --allnum --fmnist
ls ./visu/preview/[GENERATED NEW IMAGE]
| Label | Description |
|---|---|
| 0 | T-shirt/top |
| 1 | Trouser |
| 2 | Pullover |
| 3 | Dress |
| 4 | Coat |
| 5 | Sandal |
| 6 | Shirt |
| 7 | Sneaker |
| 8 | Bag |
| 9 | Ankle boot |
(model is gen_iter_18000.npz)
- T-Shirt/top
- Trouser
- pullover
- Dress
- Coat
- Sandal
- Shirt
- Sneaker
- Bag
- Ankle boot
This materials provided as MIT License(Please see LICENSE file).