CSGAN(Conditional StyleGAN) implementation with Keras. Used CelebA datasets(64 x 64 x 3) and 5 classes for training.
- tensorflow-gpu: 1.14.0
- Keras: 2.3.1
- tensorboard: 1.14.0
- h5py: 2.10.0
- numpy: 1.18.1
You need to prepare your own dataset in the dataset/ directory.
python model.py --model=CSGAN --gpu=2 --name=csgan-test --load_model=3000
There are 8 flags for training/validating parameters.
| Flags | type | default | |
|---|---|---|---|
| load_model | integer | Iteration number of the model you wish to load. '-1' for training a new model | -1 |
| validate | boolean | Set to 'True' for validating a model and 'False' for training. | False |
| glasses | boolean | When set 'True', generates 'glasses' validate images.(Only works when validate is 'True') | False |
| male | boolean | When set 'True', generates 'male' validate images. (Only works when validate is 'True') | False |
| model | enum | Select from ['CSGAN', 'ACGAN', 'CGAN']. | CSGAN |
| gpu | integer | The gpu number for training. | 0 |
| name | string | The name of your model. Will be used for saving model and validate images. | None |
| batch_size | integer | The batch size for training the model. | 64 |