/cramer_gan_chainer

Cramer GAN implemented in Chainer v3.0.0.

Primary LanguagePythonMIT LicenseMIT

This is an implementation of The Cramer Distance as a Solution to Biased Wasserstein Gradients in Chainer v3.0.0.

Requirements

Chainer v3.0.0, OpenCV, etc.
The scripts work on Python 2.7.13 and 3.6.1.

How to generate images

$ python generate_image.py example_food-101/config.py -p example_food-101/trained-params_gen_update-000040000.npz

You can generate various images by changing the random_seed option.

$ python generate_image.py example_food-101/config.py -r 1 -p example_food-101/trained-params_gen_update-000040000.npz

Example Food-101

example_image_food-101

Dataset

I resized the images to 64x64 before training.

  • Food-101
    Bossard, Lukas and Guillaumin, Matthieu and Van Gool, Luc. Food-101 -- Mining Discriminative Components with Random Forests. European Conference on Computer Vision, 2014.