GANs-TensorFlow2 is a repository that implements a variety of popular Generative Adversarial Network algorithms using TensorFlow2. The key to this repository is an easy-to-understand code. Therefore, if you are a student or a researcher studying Deep Reinforcement Learning, I think it would be the best choice to study with this repository. One algorithm relies only on one python script file. So you don't have to go in and out of different files to study specific algorithms. This repository is constantly being updated and will continue to add a new Generative Adversarial Network algorithm.
Paper Generative Adversarial Networks
Author Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, Yoshua Bengio
Publish NIPS 2014
def get_loss_fn():
def d_loss_fn(real_logits, fake_logits):
return -tf.reduce_mean(tf.math.log(real_logits + 1e-10) + tf.math.log(1. - fake_logits + 1e-10))
def g_loss_fn(fake_logits):
return -tf.reduce_mean(tf.math.log(fake_logits + 1e-10))
return d_loss_fn, g_loss_fn
$ python GAN/GAN.py
Paper Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
Author Alec Radford, Luke Metz, Soumith Chintala
Publish ICLR 2016
def get_loss_fn():
criterion = tf.keras.losses.BinaryCrossentropy(from_logits=True)
def d_loss_fn(real_logits, fake_logits):
real_loss = criterion(tf.ones_like(real_logits), real_logits)
fake_loss = criterion(tf.zeros_like(fake_logits), fake_logits)
return real_loss + fake_loss
def g_loss_fn(fake_logits):
return criterion(tf.ones_like(fake_logits), fake_logits)
return d_loss_fn, g_loss_fn
$ python DCGAN/DCGAN.py
Paper Least Squares Generative Adversarial Networks
Author Xudong Mao, Qing Li, Haoran Xie, Raymond Y.K. Lau, Zhen Wang, Stephen Paul Smolley
Publish ICCV 2017
def get_loss_fn():
criterion = tf.keras.losses.MeanSquaredError()
def d_loss_fn(real_logits, fake_logits):
real_loss = criterion(tf.ones_like(real_logits), real_logits)
fake_loss = criterion(tf.zeros_like(fake_logits), fake_logits)
return real_loss + fake_loss
def g_loss_fn(fake_logits):
return criterion(tf.ones_like(fake_logits), fake_logits)
return d_loss_fn, g_loss_fn
$ python LSGAN/LSGAN.py
Paper Wasserstein GAN
Author Martin Arjovsky, Soumith Chintala, LΓ©on Bottou
Publish arXiv 2017
def get_loss_fn():
def d_loss_fn(real_logits, fake_logits):
return tf.reduce_mean(fake_logits) - tf.reduce_mean(real_logits)
def g_loss_fn(fake_logits):
return -tf.reduce_mean(fake_logits)
return d_loss_fn, g_loss_fn
$ python WGAN/WGAN.py
Paper Improved Training of Wasserstein GANs
Author Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, Aaron Courville
Publish NIPS 2017
def get_loss_fn():
def d_loss_fn(real_logits, fake_logits):
return tf.reduce_mean(fake_logits) - tf.reduce_mean(real_logits)
def g_loss_fn(fake_logits):
return -tf.reduce_mean(fake_logits)
return d_loss_fn, g_loss_fn
def gradient_penalty(generator, real_images, fake_images):
real_images = tf.cast(real_images, tf.float32)
fake_images = tf.cast(fake_images, tf.float32)
alpha = tf.random.uniform([BATCH_SIZE, 1, 1, 1], 0., 1.)
diff = fake_images - real_images
inter = real_images + (alpha * diff)
with tf.GradientTape() as tape:
tape.watch(inter)
predictions = generator(inter)
gradients = tape.gradient(predictions, [inter])[0]
slopes = tf.sqrt(tf.reduce_sum(tf.square(gradients), axis=[1, 2, 3]))
return tf.reduce_mean((slopes - 1.) ** 2)
$ python WGAN-GP/WGAN-GP.py
Paper On Convergence and Stability of GANs
Author Naveen Kodali, Jacob Abernethy, James Hays, Zsolt Kira
Publish ICLR 2018
def get_loss_fn():
def d_loss_fn(real_logits, fake_logits):
return tf.reduce_mean(fake_logits) - tf.reduce_mean(real_logits)
def g_loss_fn(fake_logits):
return -tf.reduce_mean(fake_logits)
return d_loss_fn, g_loss_fn
def gradient_penalty(generator, real_images):
real_images = tf.cast(real_images, tf.float32)
def _interpolate(a):
beta = tf.random.uniform(tf.shape(a), 0., 1.)
b = a + 0.5 * tf.math.reduce_std(a) * beta
shape = [tf.shape(a)[0]] + [1] * (a.shape.ndims - 1)
alpha = tf.random.uniform(shape, 0., 1.)
inter = a + alpha * (b - a)
inter.set_shape(a.shape)
return inter
x = _interpolate(real_images)
with tf.GradientTape() as tape:
tape.watch(x)
predictions = generator(x, training=True)
grad = tape.gradient(predictions, x)
slopes = tf.norm(tf.reshape(grad, [tf.shape(grad)[0], -1]), axis=1)
return tf.reduce_mean((slopes - 1.) ** 2)
$ python DRAGAN/DRAGAN.py