The GANdam project

A Generative Adversarial Network (GAN) applied to a dataset of images of mechas from the anime Mobile Suit Gundam. Read the full blog post here.

Instructions

Step 1 : data scraping

cd Code/scraping/
python scrape_images_urls.py
python download_images.py

Intermediate steps are saved to the Generated/ folder.

Step 2 : preprocessing

cd Code/preprocessing/
python flatten_dataset.py

You must then create the folder Generated/preprocessing/manually_selected/ and manually copy there the images you want from Generated/preprocessing/flattened_images/.

Then :

python resize_images.py
python augment_images.py

Step 3 : training

cd Code/training/
python train.py

Checkpoints and final model are saved at Generated/training/models/, and generator samples at Generated/training/training_samples/.

Step 4 : sampling the generator

cd Code/testing/
python sample_generator.py

Samples will be saved at Generated/testing/samples/.