def gengerate(train_dir, suffix=''):
datacfg = data_cfg.get_config(args.dataset)
model = get_model()
log.info("Generating %i batches using suffix %s" %
(args.num_generated_batches, suffix))
noise = sample_generate_noise()
fake_images = model.sampler(noise, False)
init_assign_op, init_feed_dict = utils.restore_ckpt(train_dir, log)
# clean_init_op = tf.group(tf.global_variables_initializer(),
# tf.local_variables_initializer())
#saver = tf.train.Saver()
tf.get_default_graph().finalize()
with tf.Session(config=tf.ConfigProto(allow_soft_placement=True,
log_device_placement=False)) as sess:
# sess.run(clean_init_op)
#saver.restore(sess,train_dir)
sess.run(init_assign_op, feed_dict=init_feed_dict)
output_list = []
for i in range(args.num_generated_batches):
output_list.append(sess.run(fake_images))
output = np.concatenate(output_list, axis=0)
output = denormalize(output)
np.save(
os.path.join(SAMPLE, args.model_name, args.dataset,
"X_gan_%s.npy" % (args.model_name + suffix)), output)
def view_pics(args, from_disk, generated_imgs, index):
datacfg = data_cfg.get_config(args.dataset)
if from_disk:
generated_pics = np.load(
os.path.join(DATASETS, args.dataset,
'X_gan_%s.npy' % args.model_name))
else:
generated_pics = generated_imgs
gs = gridspec.GridSpec(8, 8)
gs.update(wspace=0.05, hspace=0.05)
imgs_to_show = np.array(generated_pics)
imgs_to_show = np.reshape(imgs_to_show, [
-1, datacfg.dataset.image_size * datacfg.dataset.image_size *
datacfg.dataset.channels
])
imgs_to_show = imgs_to_show[-16:]
plt.figure(figsize=(4, 4))
for i, img in enumerate(imgs_to_show):
ax = plt.subplot(gs[i])
plt.axis('off')
ax.set_xticklabels([])
ax.set_yticklabels([])
ax.set_aspect('equal')
img = np.reshape(
img, [datacfg.dataset.image_size, datacfg.dataset.image_size, 3])
plt.imshow(img)
plt.savefig('generated_pic%i.eps' % index, dpi=250)
return
def get_inception_score():
datacfg = data_cfg.get_config(args.dataset)
generated_pics = np.load(
os.path.join(SAMPLE, args.model_name, args.dataset,
'X_gan_%s.npy' % args.model_name))
#generated_pics=np.reshape(generated_pics,[-1,datacfg.dataset.image_size,datacfg.dataset.image_size,datacfg.dataset.channels])
mean, var = inception_score.get_inception_score(generated_pics)
log.info('the inception score is %s,with d=standard deviation %s' %
(mean, var))
def construct_model():
model=get_model()
real_images,iter_fn=matcher.load_dataset('train',args.batch_size,args.dataset,32)
datacfg=data_cfg.get_config(args.dataset)
model.construct_variables(args.noise_dim,datacfg)
input_noise=sample_noise()
fake_img=model.get_generator(input_noise,datacfg)
logits_real=model.get_discriminator(real_images,datacfg)
logits_fake=model.get_discriminator(fake_img,datacfg)
optimizer=get_optimizer(name="",optimizer=args.optimizer)
D_loss,G_Loss=model.get_loss(logits_real,logits_fake)
D_vars=tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES,'discriminator')
G_vars=tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES,'generator')
Discriminator_train_op=optimizer.minimize(D_loss,var_list=D_vars)
Generator_train_op=optimizer.minimize(G_Loss,var_list=G_vars)
return Discriminator_train_op,Generator_train_op,iter_fn,D_loss,G_Loss,fake_img
import logging
from easydict import EasyDict as edict
from data_process import matcher
from model import origin_dcgan
import data_config as data_cfg
from evaluation import inception_score
from evaluation import frachet_inception_distance
import viewpics
from viewpics import image_manifold_size
from model import deep_dcgan
from model import dcgan_standard
sample_dir = os.path.join('./samples', args.model_name, args.dataset)
logging.config.dictConfig(get_logging_config(args.model_name))
log = logging.getLogger("gan")
data_config = data_cfg.get_config(args.dataset)
def sample_noise():
return tf.random_uniform([args.batch_size, args.noise_dim],
minval=-1,
maxval=1)
def sample_generate_noise():
return tf.random_uniform([100, args.noise_dim], minval=-1, maxval=1)
def get_optimizer(name, optimizer=args.optimizer):
if args.lr_decay:
global_step = tf.train.get_global_step()