try:
os.makedirs(IMAGES_DIR)
except:
pass
# placeholders for data going into the network
global_step = tf.Variable(0, name='global_step', trainable=False)
z = tf.placeholder(tf.float32, shape=(BATCH_SIZE, 100), name='z')
train_images_list = data_ops.loadData(DATA_DIR, DATASET)
filename_queue = tf.train.string_input_producer(train_images_list)
real_images = data_ops.read_input_queue(filename_queue, BATCH_SIZE)
# generated images
gen_images = netG(z, BATCH_SIZE)
# get the output from D on the real and fake data
errD_real = netD(real_images, BATCH_SIZE, SELU, NORM)
errD_fake = netD(gen_images, BATCH_SIZE, SELU, NORM, reuse=True)
# cost functions
errD = tf.reduce_mean(errD_real) - tf.reduce_mean(errD_fake)
errG = tf.reduce_mean(errD_fake)
# gradient penalty
epsilon = tf.random_uniform([], 0.0, 1.0)
x_hat = real_images * epsilon + (1 - epsilon) * gen_images
d_hat = netD(x_hat, BATCH_SIZE, SELU, NORM, reuse=True)
gradients = tf.gradients(d_hat, x_hat)[0]
slopes = tf.sqrt(tf.reduce_sum(tf.square(gradients),
checkpoint_dir = sys.argv[1]
n = 15 # cols
m = 5 # rows
num_images = n * m
img_size = (112, 112, 3)
canvas = 255 * np.ones(
(m * img_size[0] + (10 * m) + 10, n * img_size[1] + (10 * n) + 10, 3),
dtype=np.uint8)
z = tf.placeholder(tf.float32, shape=(num_images, 1024), name='z')
generated_images = netG(z, num_images)
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(checkpoint_dir)
if ckpt and ckpt.model_checkpoint_path:
print "Restoring previous model..."
try:
saver.restore(sess, ckpt.model_checkpoint_path)
print "Model restored"
except:
raise
print "Could not restore model"
# placeholders for data going into the network
global_step = tf.Variable(0, name='global_step', trainable=False)
z1 = tf.placeholder(tf.float32, shape=(BATCH_SIZE, 100), name='z1')
z2 = tf.placeholder(tf.float32, shape=(BATCH_SIZE, 100), name='z2')
train_images_list = data_ops.loadData(DATA_DIR, DATASET)
filename_queue = tf.train.string_input_producer(train_images_list)
# sample from true data
real_images = data_ops.read_input_queue(filename_queue, BATCH_SIZE)
# dummy to initialize D
dummy = netD(real_images, reuse=False)
# sample two independent images from the generator
gen_images1 = netG(z1, BATCH_SIZE)
gen_images2 = netG(z2, BATCH_SIZE, reuse=True)
# define the critic
def critic(x):
return tf.norm(netD(x, reuse=True) - netD(gen_images2, reuse=True),
axis=1) - tf.norm(netD(x, reuse=True), axis=1)
# sample epsilon from uniform distribution
epsilon = tf.random_uniform([], 0.0, 1.0)
# interpolate real and generated first samples
x_hat = epsilon * real_images + (1 - epsilon) * gen_images1
d_hat = critic(x_hat)
ddx = tf.gradients(d_hat, x_hat)[0]
ddx = tf.norm(ddx, axis=1)