def save_vis_masks(ims,im_names,gt_masks_dir,gt_masks,corrup_masks_dir,corrup_masks):
# 5.2> save vis_gt_masks and corrup_masks
alpha_folder='alpha_masks'
color_folder='color_masks'
## alpha masks(gt)
alpha_gt_dir=os.path.join(gt_masks_dir,alpha_folder)
create_dir(alpha_gt_dir)
alpha_gt_masks=gt_masks*255.0
save_images(alpha_gt_masks,alpha_gt_dir,im_names,im_ext)
## alpha masks(corrup)
alpha_corrup_dir=os.path.join(corrup_masks_dir,alpha_folder)
create_dir(alpha_corrup_dir)
alpha_corrup_masks=corrup_masks*255.0
save_images(alpha_corrup_masks,alpha_corrup_dir,im_names,im_ext)
## color masks(gt)
color_gt_masks=np.zeros((ims.shape),dtype=np.uint8)
color=[0,0,255]
for m_id in xrange(len(gt_masks)):
im=ims[m_id]
mask=gt_masks[m_id]
color_mask=vis_im_mask(im,mask,color,True)
color_gt_masks[m_id]=color_mask
color_gt_dir=os.path.join(gt_masks_dir,color_folder)
create_dir(color_gt_dir)
save_images(color_gt_masks,color_gt_dir,im_names,im_ext)
## here is small bug(need to debug)
## color masks(corrup)
color_corrup_masks=np.zeros((ims.shape),dtype=np.uint8)
color=[0,0,255]
for m_id in xrange(len(corrup_masks)):
im=ims[m_id]
mask=corrup_masks[m_id]
color_mask=vis_im_mask(im,mask,color,True)
color_corrup_masks[m_id]=color_mask
color_corrup_dir=os.path.join(corrup_masks_dir,color_folder)
create_dir(color_corrup_dir)
save_images(color_corrup_masks,color_corrup_dir,im_names,im_ext)
dcgan.input_label: one_label
})
d_loss = d_loss_fake + d_loss_real
g_loss = g_loss1 + g_loss2
writer.add_summary(ds, epoch * batch_count + i)
writer.add_summary(gs, epoch * batch_count + i)
print(
'EPOCH {}'.format(epoch),
'[{}/{}] D_Loss: {}, G_Loss: {}'.format(
i, batch_count, d_loss, g_loss))
if i % save_image_interval == 0:
# de-preprocess image & save
z = np.random.normal(loc=0.0,
scale=0.1,
size=[batch_size, 100])
gen = sess.run(dcgan.generated,
feed_dict={dcgan.input_noise: z})
if channel_num > 1:
gen = np.reshape(gen, [batch_size, 64, 64, channel_num])
else:
gen = np.reshape(gen, [batch_size, 64, 64])
gen_img = gen + 1.0 * 127.0
if not os.path.exists('gen_celeb'):
os.makedirs('gen_celeb')
image_util.save_images('gen_celeb/{}_{}.png'.format(epoch, i),
gen_img[0:64], [8, 8])
self.loss_sum = tf.summary.scalar('loss', self.loss)
self.sum = tf.summary.merge([self.loss_sum])
#
# hyper-parameter
#
batch_size = 200
dcgan = DCGAN(batch_size=batch_size, channel=1)
# load MNIST data
images, labels = mnist_data.load_mnist('./mnist')
input_img = []
for i in range(60000):
input_img.append(images[i])
input_img = np.array(input_img)
image_util.save_images('output.png', input_img[0:64], [8,8])
# preprocess images
input_img = input_img / 127.0 - 1.0
input_img = np.reshape(input_img, [60000, 28, 28, 1])
# init TF
init = tf.global_variables_initializer()
saver = tf.train.Saver()
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
with tf.Session(config=config) as sess:
sess.run(init)
writer = tf.summary.FileWriter('./graphs_mnist', sess.graph)
import os
import numpy as np
from image_util import save_image, save_images
def load_mnist(path):
fd = open(os.path.join(path, 'train-images-idx3-ubyte'))
images = np.fromfile(file=fd, dtype=np.uint8)
images = images[16:].reshape([60000, 28, 28]).astype(np.float)
fd = open(os.path.join(path, 'train-labels-idx1-ubyte'))
labels = np.fromfile(file=fd, dtype=np.uint8)
labels = labels[8:].reshape([60000]).astype(np.float)
return images, labels
# TEST DRIVE
images, labels = load_mnist('./mnist')
save_images('output.png', images[0:64], [8, 8])