def main(_):
if not tf.gfile.Exists(FLAGS.logdir):
tf.gfile.MakeDirs(FLAGS.logdir)
logging.set_verbosity(logging.INFO)
with tf.Session() as sess:
if FLAGS.similarity:
dataset = zap_data(FLAGS, False)
all_features, all_paths = export_intermediate(FLAGS, sess, dataset)
similarity(FLAGS, sess, all_features, all_paths)
else:
dataset = zap_data(FLAGS, True)
gan(dataset, sess)
def main(_):
if not tf.gfile.Exists(FLAGS.logdir):
tf.gfile.MakeDirs(FLAGS.logdir)
if FLAGS.sampledir and not tf.gfile.Exists(FLAGS.sampledir):
tf.gfile.MakeDirs(FLAGS.sampledir)
with tf.Session() as sess:
if FLAGS.sampledir:
sample(FLAGS, sess)
elif FLAGS.similarity:
dataset = zap_data(FLAGS, False)
all_features, all_paths = export_intermediate(FLAGS, sess, dataset)
data_json = similarity(FLAGS, sess, all_features, all_paths)
similarity_view(data_json)
else:
dataset = zap_data(FLAGS, True)
gan(dataset, sess)
def gan(cluster):
# Model
is_chief = (FLAGS.task == 0)
local_step = 0
server = tf.train.Server(cluster, job_name='worker', task_index=FLAGS.task)
worker_device = '/job:worker/task:%d' % (FLAGS.task)
with tf.device(
tf.train.replica_device_setter(worker_device=worker_device,
ps_device='/job:ps',
cluster=cluster)), tf.Session():
dataset = zap_data(FLAGS, True)
num_global = (dataset['size'] / FLAGS.batch_size) * FLAGS.epochs
x = tf.placeholder(tf.float32,
shape=[
None, IMAGE_SIZE['resized'][0],
IMAGE_SIZE['resized'][1], 3
])
dropout = tf.placeholder(tf.float32)
d_model = discriminator(x, reuse=False, dropout=dropout)
z = tf.placeholder(tf.float32, shape=[None, Z_DIM])
latent_c = tf.placeholder(shape=[None, C_DIM], dtype=tf.float32)
g_model = generator(z, latent_c)
dg_model, q_model = discriminator(g_model,
reuse=True,
dropout=dropout,
c_dim=C_DIM)
d_trainer, d_loss, g_trainer, g_loss, global_step = loss(
d_model, g_model, dg_model, q_model, latent_c)
# Stats
t_vars = tf.trainable_variables()
count_params(t_vars, ['discr/', 'gen/', 'latent_c/'])
# for v in t_vars:
# tf.histogram_summary(v.name, v)
# Init
summary = tf.summary.merge_all()
init_op = tf.global_variables_initializer()
sv = tf.train.Supervisor(is_chief=is_chief,
logdir=FLAGS.logdir,
init_op=init_op,
recovery_wait_secs=1,
summary_op=None,
global_step=global_step)
sess_config = tf.ConfigProto(
allow_soft_placement=True,
log_device_placement=False,
device_filters=['/job:ps',
'/job:worker/task:%d' % FLAGS.task])
step = 0
with sv.managed_session(server.target, config=sess_config) as sess:
# Dataset queue
#coord = tf.train.Coordinator()
#threads = tf.train.start_queue_runners(coord=coord)
#tf.train.start_queue_runners(sess=sess)
while step < num_global and not sv.should_stop():
z_batch = np.random.uniform(
-1, 1, [FLAGS.batch_size, Z_DIM]).astype(np.float32)
c_batch = np.random.uniform(-1, 1, [FLAGS.batch_size, C_DIM])
images, _ = sess.run(dataset['batch'])
feed_dict = {
z: z_batch,
latent_c: c_batch,
x: images,
dropout: .5,
}
# Update discriminator
start = time.time()
_, d_loss_val = sess.run([d_trainer, d_loss],
feed_dict=feed_dict)
d_time = time.time() - start
# Update generator
start = time.time()
_, g_loss_val, summary_str, step = sess.run(
[g_trainer, g_loss, summary, global_step],
feed_dict=feed_dict)
g_time = time.time() - start
# Log details
if local_step % 10 == 0:
print(
"[%s, %s] Disc loss: %.3f (%.2fs), Gen Loss: %.3f (%.2fs)"
% (
step,
step * FLAGS.batch_size / dataset['size'],
d_loss_val,
d_time,
g_loss_val,
g_time,
))
if is_chief:
sv.summary_computed(sess, summary_str)
local_step += 1
# Early stopping
if np.isnan(g_loss_val) or np.isnan(d_loss_val):
print('Early stopping', g_loss_val, d_loss_val)
break
# Finish off the filename queue coordinator.
#coord.request_stop()
#coord.join(threads)
sv.request_stop()
return
import tensorflow as tf
from main import discriminator
from zap50k import zap_data, IMAGE_SIZE
import itertools
import scipy.misc
import pickle
import flask
flags = tf.app.flags
FLAGS = flags.FLAGS
app = flask.Flask(__name__)
sess = tf.Session()
with sess.as_default():
dataset = zap_data(FLAGS, False)
x = tf.placeholder(
tf.float32,
shape=[None, IMAGE_SIZE['resized'][0], IMAGE_SIZE['resized'][1], 3])
dropout = tf.placeholder(tf.float32)
feat_model = discriminator(x, reuse=False, dropout=dropout, int_feats=True)
all_features = np.zeros((dataset['size'], feat_model.get_shape()[1]))
x1 = tf.placeholder(tf.float32, shape=[None, all_features.shape[1]])
x2 = tf.placeholder(tf.float32, shape=[None, all_features.shape[1]])
l2diff = tf.sqrt(tf.reduce_sum(tf.square(tf.subtract(x1, x2)), axis=1))
# Init
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)