Progressive GAN trained on CelebA for 128x128 images.
Module URL: https://tfhub.dev/google/progan-128/1
Image generator based on tensorflow reimplementation of Progressive GANs.
Maps from a 512-dimensional latent space to images. During training, the latent space vectors were sampled from a normal distribution.
<Tensor(tf.float32, shape=[?, 512])>, representing a batch of
latent vectors as input, and outputs
<Tensor(tf.float32, shape=[?, 128, 128, 3])> representing a batch of RGB
# Generate 20 random samples. generate = hub.Module("https://tfhub.dev/google/progan-128/1") images = generate(tf.random_normal([20, 512]))
The original model has been trained on a GPU for 636,801 steps with batch size 16.
 Tero Karras, Timo Aila, Samuli Laine, Jaakko Lehtinen. Progressive Growing of GANs for Improved Quality, Stability, and Variation. arXiv preprint arXiv:1710.10196, 2017.