This project consists of three parts: STGAN, MTGAN and SGAN
SGAN is written in tensorflow, but the model structure has some references to the original Theano implementation:
https://github.com/zalandoresearch/spatial_gan
Except the OS modules, image processing functions and the flags set-ups, most codes are written by myself.
To use SGAN, simpy run python run.py
You must have a directory called dataset that has all training samples in it.
To change parameters, use -- For example, python run.py --spatial_size=7
To generate images from pretrained models, use python run.py --training=False --model_dir=""
Here are some good samples:
