/tf_dcgan

A TensorFlow implementation of dcgan(Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks)

Primary LanguagePythonMIT LicenseMIT

tf_dcgan

A TensorFlow implementation of DC-GAN:

Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks

Dataset

We will use celebA dataset in this implementation, please download it here

Unzip the file, then

ls *.png > img_set.txt

And then delete the last line (which is the name of this txt file itself)

Some Generated Examples (fixed noise)

(I used the wrong color channel order to save, it is actually BGR now)

Misc

Both generator loss and discriminator loss become nan in about 13th epoch. Not yet figured out why.

Reference

THe PyTorch example