Pre-trained generators v1.0

This release contains pytorch checkpoints for the generators trained using OCFGAN-GP. These generators can be used to generate new image samples.

To generate an 8x8 grid of image samples, run the following command:

python src/gen_samples.py\
 --png\
 --imsize <image size>\
 --noise_dim <noise dim>\
 --gen <generator network>\
 --out_dir ./ <checkpoint>.pth

Omit the --png flag to save the samples to a numpy file.

The command line arguments for the different datasets are as follows:

• CelebA

--imsize 32 --noise_dim 64 --gen flexible-dcgan

• CelebA-128

--imsize 128 --noise_dim 100 --gen resnet

• CIFAR10

--imsize 32 --noise_dim 32 --gen flexible-dcgan

• STL10

--imsize 32 --noise_dim 32 --gen flexible-dcgan