print(psnr_vals_list)
# -------------------------------- Train loop ---------------------------------------------
config = tf.ConfigProto() # dynamic memory growth
config.gpu_options.allow_growth = True
with tf.Session(config=config) as session:
# Init variables
if (START_ITER > 0):
saver.restore(session,
restore_path) # Restore variables from saved session.
print("Model restored.")
plotter.restore(START_ITER) # makes plots start from 0
session.run(fixed_noise.initializer)
else:
session.run(init_op)
session.run(fixed_noise.initializer)
print(fixed_noise.eval())
overall_start_time = time.time()
summary_writer = tf.summary.FileWriter(log_dir, graph=session.graph)
# Network Training
for iteration in range(START_ITER,
ITERS): # START_ITER: 0 or from last checkpoint
start_time = time.time()
imsaver.save_images(samples_255.reshape((BATCH_SIZE, 3, IM_DIM, IM_DIM)),
'samples_{}.png'.format(frame)) ### for MNIST
init_op = tf.global_variables_initializer() # op to initialize the variables.
saver = tf.train.Saver() # ops to save and restore all the variables.
# Train loop
with tf.Session() as session:
# Init variables
if (CONTINUE):
saver.restore(session,
restore_path) # Restore variables from saved session.
print("Model restored.")
plotter.restore(
START_ITER
) # does not fully work, but makes plots start from newly started iteration
else:
session.run(init_op)
for iteration in range(START_ITER,
ITERS): # START_ITER: 0 or from last checkpoint
start_time = time.time()
# Train generator
if iteration > 0:
_ = session.run(gen_train_op)
# Train duscriminator
for i in range(DISC_ITERS):
_data, _ = next(gen) # shape: (batchsize, 6144)
_real_data = _data[:, 2 * OUTPUT_DIM:] # current frame for disc
_disc_cost, _ = session.run([disc_cost, disc_train_op],