Implementation of Improved Training of Wasserstein GANs (Gulrajani, et al.) in PyTorch
PyTorch
Numpy
scipy
(for loading SVHN .mat file)
Run the following code to train on the SVHN dataset
python code/train_SVHN.py <location-of-SVHN-mat-file> <model-name>
The parameters can be changed through the command line. Use the --help
flag for more details
This image compares real data and data generated from the model This image demonstrates an interpolation from two points in the noise space of the generator. A straight line was drawn between two randomly drawn vectors and the points along that line were fed into the generator to produce these images.