- Python 3.3+
- Tensorflow 0.12.1
- SciPy
- pillow
- tqdm
- (Optional) moviepy (for visualization)
- (Optional) Align&Cropped Images.zip : Large-scale CelebFaces Dataset
First, download MNIST dataset with:
$ python3 download.py mnist
To train a model with downloaded dataset:
$ python3 main.py --dataset mnist --input_height=28 --output_height=28 --c_dim=1 --is_train
To test with an existing model:
$ python3 main.py --dataset mnist --input_height=28 --output_height=28 --c_dim=1
This Mix-DCGAN repository was build off Taehoon Kim's / @carpedm20 DCGAN implementation. The MIX-GAN is based off of the paper written in: Generalization and Equilibrium in Generative Adversarial Nets
- T=5 set currently
- Change self.T in model.py to set number of parallel Generators and Discriminators
Kenny Gea, Logan Engstrom