def display_fig(name, figure_A, figure_B):
figure = np.concatenate([figure_A, figure_B], axis=0 )
columns = 4
elements = figure.shape[0]
figure = figure.reshape((columns,(elements//columns)) + figure.shape[1:])
figure = stack_images(figure)
figure = np.clip((figure + 1) * 255 / 2, 0, 255).astype('uint8')
display_fn(figure, name)
def show_sample(self):
""" Display preview data """
if len(self.images) != 2:
logger.debug("Ping Pong training - Only one side trained. Aborting preview")
return None
logger.debug("Showing sample")
feeds = dict()
figures = dict()
headers = dict()
for side, samples in self.images.items():
faces = samples[1]
if self.model.input_shape[0] / faces.shape[1] != 1.0:
feeds[side] = self.resize_sample(side, faces, self.model.input_shape[0])
feeds[side] = feeds[side].reshape((-1, ) + self.model.input_shape)
else:
feeds[side] = faces
if self.use_mask:
mask = samples[-1]
feeds[side] = [feeds[side], mask]
preds = self.get_predictions(feeds["a"], feeds["b"])
for side, samples in self.images.items():
other_side = "a" if side == "b" else "b"
predictions = [preds["{}_{}".format(side, side)],
preds["{}_{}".format(other_side, side)]]
display = self.to_full_frame(side, samples, predictions)
headers[side] = self.get_headers(side, other_side, display[0].shape[1])
figures[side] = np.stack([display[0], display[1], display[2], ], axis=1)
if self.images[side][0].shape[0] % 2 == 1:
figures[side] = np.concatenate([figures[side],
np.expand_dims(figures[side][0], 0)])
width = 4
side_cols = width // 2
if side_cols != 1:
headers = self.duplicate_headers(headers, side_cols)
header = np.concatenate([headers["a"], headers["b"]], axis=1)
figure = np.concatenate([figures["a"], figures["b"]], axis=0)
height = int(figure.shape[0] / width)
figure = figure.reshape((width, height) + figure.shape[1:])
figure = stack_images(figure)
figure = np.vstack((header, figure))
logger.debug("Compiled sample")
return np.clip(figure * 255, 0, 255).astype('uint8')
def show_sample(self):
""" Display preview data """
if len(self.images) < 2:
logger.debug(
"Ping Pong training - Only one side trained. Aborting preview")
return None
logger.debug("Showing sample")
feeds = dict()
figures = dict()
headers = dict()
for side, samples in self.images.items():
side = side[-1]
faces = samples[1]
if self.model.input_shape[0] / faces.shape[1] != 1.0:
feeds[side] = self.resize_sample(side, faces,
self.model.input_shape[0])
feeds[side] = feeds[side].reshape((-1, ) +
self.model.input_shape)
else:
feeds[side] = faces
preds = self.get_predictions(feeds)
for side, samples in self.images.items():
predictions = [
preds["{}_{}".format(other_side, side[-1])]
for other_side in range(len(feeds))
]
#print(f"------predictions len = {len(predictions)}")
display = self.to_full_frame(side, samples, predictions)
#headers[side] = self.get_headers(side, len(feeds), display[0].shape[1])
#print(f"-----------display len = {len(display)}")
figures[side] = np.stack(display, axis=1)
width = len(feeds) + 1
#header = np.concatenate([headers["0"], headers["1"]], axis=1)
#print(f"---------------------------------figures len = {figures['1'].shape}")
figure = np.concatenate([figures[key] for key in figures], axis=1)
#print(f"-----------------------------------{figure.shape}")
height = len(feeds)
figure = figure.reshape((height, width) + figure.shape[2:])
figure = stack_images(figure)
#figure = np.vstack((header, figure))
logger.debug("Compiled sample")
return np.clip(figure * 255, 0, 255).astype('uint8')
def show_sample(self, test_A, test_B):
figure_A = numpy.stack([
test_A,
self.model.autoencoder_A.predict(test_A),
self.model.autoencoder_B.predict(test_A),
], axis=1)
figure_B = numpy.stack([
test_B,
self.model.autoencoder_B.predict(test_B),
self.model.autoencoder_A.predict(test_B),
], axis=1)
figure = numpy.concatenate([figure_A, figure_B], axis=0)
figure = figure.reshape((4, 7) + figure.shape[1:])
figure = stack_images(figure)
return numpy.clip(figure * 255, 0, 255).astype('uint8')
def show_sample(self, test_A, test_B):
figure_A = numpy.stack([
test_A,
self.model.autoencoder_A.predict(test_A),
self.model.autoencoder_B.predict(test_A),
], axis=1)
figure_B = numpy.stack([
test_B,
self.model.autoencoder_B.predict(test_B),
self.model.autoencoder_A.predict(test_B),
], axis=1)
figure = numpy.concatenate([figure_A, figure_B], axis=0)
figure = figure.reshape((4, 7) + figure.shape[1:])
figure = stack_images(figure)
return numpy.clip(figure * 255, 0, 255).astype('uint8')
def showG_mask(self, test_A, test_B, path_A, path_B):
figure_A = np.stack([
test_A,
(np.squeeze(np.array([path_A([test_A[i:i+1]]) for i in range(test_A.shape[0])])))*2-1,
(np.squeeze(np.array([path_B([test_A[i:i+1]]) for i in range(test_A.shape[0])])))*2-1,
], axis=1 )
figure_B = np.stack([
test_B,
(np.squeeze(np.array([path_B([test_B[i:i+1]]) for i in range(test_B.shape[0])])))*2-1,
(np.squeeze(np.array([path_A([test_B[i:i+1]]) for i in range(test_B.shape[0])])))*2-1,
], axis=1 )
figure = np.concatenate([figure_A, figure_B], axis=0 )
figure = figure.reshape((4,self.batch_size // 2) + figure.shape[1:])
figure = stack_images(figure)
figure = np.clip((figure + 1) * 255 / 2, 0, 255).astype('uint8')
return figure
def show_sample(self, test_A, test_B):
AA = self.model.autoencoder_A.predict(test_A)
BA = self.model.autoencoder_B.predict(test_A)
BB = self.model.autoencoder_B.predict(test_B)
AB = self.model.autoencoder_A.predict(test_B)
figure_A = numpy.stack(
[
test_A,
#self.model.autoencoder_A.predict(test_A),
AA,
#self.model.autoencoder_B.predict(test_A),
BA,
],
axis=1)
figure_B = numpy.stack(
[
test_B,
#self.model.autoencoder_B.predict(test_B),
BB,
#self.model.autoencoder_A.predict(test_B),
AB,
],
axis=1)
if (test_A.shape[0] % 2) != 0:
figure_A = numpy.concatenate(
[figure_A, numpy.expand_dims(figure_A[0], 0)])
figure_B = numpy.concatenate(
[figure_B, numpy.expand_dims(figure_B[0], 0)])
figure = numpy.concatenate([figure_A, figure_B], axis=0)
w = 4
h = int(figure.shape[0] / w)
figure = figure.reshape((w, h) + figure.shape[1:])
figure = stack_images(figure)
return numpy.clip(figure * 255, 0, 255).astype('uint8')
def show_sample(self, test_A, test_B):
figure_A = numpy.stack([
test_A,
autoencoder_A.predict(test_A),
autoencoder_B.predict(test_A),
], axis=1)
figure_B = numpy.stack([
test_B,
autoencoder_B.predict(test_B),
autoencoder_A.predict(test_B),
], axis=1)
figure = numpy.concatenate([figure_A, figure_B], axis=0)
figure = figure.reshape((4, 7) + figure.shape[1:])
figure = stack_images(figure)
figure = numpy.clip(figure * 255, 0, 255).astype('uint8')
if self.arguments.preview is True:
cv2.imshow('', figure)
if not self.arguments.preview or self.arguments.write_image:
cv2.imwrite('_sample.jpg', figure)
def show_sample(self, test_A, test_B):
figure_A = numpy.stack([
test_A,
self.model.autoencoder_A.predict(test_A),
self.model.autoencoder_B.predict(test_A),
], axis=1)
figure_B = numpy.stack([
test_B,
self.model.autoencoder_B.predict(test_B),
self.model.autoencoder_A.predict(test_B),
], axis=1)
if test_A.shape[0] % 2 == 1:
figure_A = numpy.concatenate ([figure_A, numpy.expand_dims(figure_A[0],0) ])
figure_B = numpy.concatenate ([figure_B, numpy.expand_dims(figure_B[0],0) ])
figure = numpy.concatenate([figure_A, figure_B], axis=0)
w = 4
h = int( figure.shape[0] / w)
figure = figure.reshape((w, h) + figure.shape[1:])
figure = stack_images(figure)
return numpy.clip(figure * 255, 0, 255).astype('uint8')
