/gan

Simple GAN with keras and TensorFlow, can be trained on mnist and cifar10

Primary LanguagePython

Simple GAN

GAN via Keras and TensorFlow

How to run

You need to have conda environment named env with keras, tensorflow, matplotlib, scipy, numpy and other packages installed. All scripts are given (or exist in repo) assuming this environment name.

Training

# Starts training and logs output into log.txt
bash run

Modify parameters inside __init__.py to achieve needed results. Datasets can be switched there too. Net snapshots are saved to ./weights/ subdirectory, also specified in that file.

Loading Pretrained and Generating

. activate env
python run_loaded.py

This will load pretrained model and generate images of each class into specified files. Modify parameters inside to control the output.