/wgan_gp_chainer

Wasserstein GAN with gradient penalty (WGAN-GP) implemented in Chainer v3.0.0.

Primary LanguagePythonMIT LicenseMIT

This is an implementation of Improved Training of Wasserstein GANs 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 fixed images by specifying 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

Example Birds

example_image_birds

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.
  • Birds
    Svetlana Lazebnik, Cordelia Schmid, and Jean Ponce. A Maximum Entropy Framework for Part-Based Texture and Object Recognition. Proceedings of the IEEE International Conference on Computer Vision, Beijing, China, October 2005, vol. 1, pp. 832-838.