print("Starting training...")
num_epochs = 200
gen_label = numpy.zeros(100).astype(numpy.int32)
adv_label = numpy.zeros(200).astype(numpy.int32)
adv_label[100:200] = 1
for epoch in range(num_epochs):
# In each epoch, we do a full pass over the training data:
train_err = 0
adv_err = 0
train_batches = 0
start_time = time.time()
for batch in mini_batch.iterate_minibatches_pair(flower_corrupt_train,
flower_truth_train,
100,
shuffle=True):
if epoch < 50:
inputs, targets = batch
train_err += train_fn(inputs, targets)
train_batches += 1
else:
inputs, targets = batch
err, gen_pic = generator_fn(inputs, targets, gen_label)
train_err += err
adv_feed = numpy.concatenate((gen_pic, targets), axis=0)
adv_err += discriminator_fn(adv_feed, adv_label)
train_batches += 1
# compute the loss on test set
view_fn = theano.function([input_var], test_seg)
# The training loop
print("Starting training...")
num_epochs = 50
for epoch in range(num_epochs):
# In each epoch, we do a full pass over the training data:
train_err = 0
train_batches = 0
start_time = time.time()
for batch in mini_batch.iterate_minibatches_pair(flower_train,
segment_train,
100,
shuffle=True):
inputs, targets = batch
train_err += train_fn(inputs, targets)
train_batches += 1
# Then we print the results for this epoch:
print("Epoch {} of {} took {:.3f}s".format(epoch + 1, num_epochs,
time.time() - start_time))
print(" training loss:\t\t{:.6f}".format(train_err / train_batches))
################### View reconstruction###########