inputs = Input(shape=(256, 256, 3))
transform_net = TransformNet(inputs, options.conv_filters, options.num_resids)
model = Model(inputs=inputs, outputs=transform_net)
loss_fn = create_loss_fn(style_target, options.content_weight,
options.style_weight, options.tv_weight,
options.batch_size)
model.compile(optimizer='adam', loss=loss_fn)
if options.model_input:
model.load_weights(options.model_input)
gen = create_gen(options.train_path,
target_size=(256, 256),
batch_size=options.batch_size)
if options.steps_per_epoch is None:
num_samples = count_num_samples(options.train_path)
options.steps_per_epoch = num_samples // options.batch_size
callbacks = None
if options.test:
callbacks = [
OutputPreview(options.test, options.test_increment, options.test_dir)
]
model.fit_generator(gen,
steps_per_epoch=options.steps_per_epoch,
epochs=options.epochs,
callbacks=callbacks)
model.save(options.model_output)
self.increment = increment
self.iteration = 0
def on_batch_end(self, batch, logs={}):
if (self.iteration % self.increment == 0):
output_img = self.model.predict(self.test_img)[0]
fname = '%d.jpg' % self.iteration
out_path = os.path.join(self.preview_dir_path, fname)
imsave(out_path, output_img)
self.iteration += 1
gen = create_gen(TRAIN_PATH, TARGET_SIZE, BATCH_SIZE)
num_samples = count_num_samples(TRAIN_PATH)
steps_per_epoch = num_samples // BATCH_SIZE
target_layer = 1
encoder_decoder = EncoderDecoder(target_layer=target_layer)
callbacks = [
OutputPreview(
encoder_decoder,
r'C:\Users\MillerV\Documents\Masters CS\Machine-Learning-Group\advancedML\lab2\monkey.jpg',
5000, './preview-%d' % target_layer)
]
encoder_decoder.model.fit_generator(gen,
steps_per_epoch=steps_per_epoch,
epochs=epochs,