def _forward_pass(self, file_path):
lr_img = imageio.imread(file_path)
if lr_img.shape[2] == 3:
lr_img = process_array(lr_img)
sr_img = self.model.model.predict(lr_img)
sr_img = process_output(sr_img)
return sr_img
else:
self.logger.error('{} is not an image with 3 channels.'.format(file_path))
def predict(self, input_image_array, by_patch_of_size=None, batch_size=10, padding_size=2):
"""
Processes the image array into a suitable format
and transforms the network output in a suitable image format.
Args:
input_image_array: input image array.
by_patch_of_size: for large image inference. Splits the image into
patches of the given size.
padding_size: for large image inference. Padding between the patches.
Increase the value if there is seamlines.
batch_size: for large image inferce. Number of patches processed at a time.
Keep low and increase by_patch_of_size instead.
Returns:
sr_img: image output.
"""
if by_patch_of_size:
lr_img = process_array(input_image_array, expand=False)
patches, p_shape = split_image_into_overlapping_patches(
lr_img, patch_size=by_patch_of_size, padding_size=padding_size
)
# return patches
mag_num = int(np.log2(self.scale))
collect={}
for i in range(0, len(patches), batch_size):
batch = self.model.predict(patches[i: i + batch_size])
if i == 0:
for j in range(mag_num):
collect[j] = batch[j]
else:
for j in range(mag_num):
collect[j] = np.append(collect[j], batch[j], axis=0)
scale=2
sr_img={}
for i in range(mag_num):
padded_size_scaled = tuple(np.multiply(p_shape[0:2], int(np.exp2(i+1)))) + (3,)
scaled_image_shape = tuple(np.multiply(input_image_array.shape[0:2], int(np.exp2(i+1)))) + (3,)
sr_img[i] = stich_together(
collect[i],
padded_image_shape=padded_size_scaled,
target_shape=scaled_image_shape,
padding_size=padding_size * int(np.exp2(i+1))
)
else:
lr_img = process_array(input_image_array)
sr_img = self.model.predict(lr_img)[0]
for i in range(mag_num):
sr_img[i] = process_output(sr_img[i])
return sr_img
def predict(self, input_image_array):
"""
Processes the image array into a suitable format
and transforms the network output in a suitable image format.
Args:
input_image_array: input image array.
Returns:
sr_img: image output.
"""
lr_img = process_array(input_image_array)
sr_img = self.model.predict(lr_img)
sr_img = process_output(sr_img)
return sr_img
def _predict(model,
input_image_array,
by_patch_of_size=None,
batch_size=10,
padding_size=2):
"""
Processes the image array into a suitable format
and transforms the network output in a suitable image format.
Args:
input_image_array: input image array.
by_patch_of_size: for large image inference. Splits the image into
patches of the given size.
padding_size: for large image inference. Padding between the patches.
Increase the value if there is seamlines.
batch_size: for large image inferce. Number of patches processed at a time.
Keep low and increase by_patch_of_size instead.
Returns:
sr_img: image output.
"""
if by_patch_of_size:
self.logger.info("Patches of size {}".format(by_patch_of_size))
lr_img = process_array(input_image_array, expand=False)
patches, p_shape = split_image_into_overlapping_patches(
lr_img, patch_size=by_patch_of_size, padding_size=padding_size)
# return patches
self.logger.info("No. patches: {}".format(len(patches)))
for i in range(0, len(patches), batch_size):
self.logger.info("Processing patch: {}/{}".format(
i, len(patches)))
batch = model.predict(patches[i:i + batch_size])
if i == 0:
collect = batch
else:
collect = np.append(collect, batch, axis=0)
scale = model.scale
padded_size_scaled = tuple(np.multiply(p_shape[0:2],
scale)) + (3, )
scaled_image_shape = tuple(
np.multiply(input_image_array.shape[0:2], scale)) + (3, )
sr_img = stich_together(
collect,
padded_image_shape=padded_size_scaled,
target_shape=scaled_image_shape,
padding_size=padding_size * scale,
)
sr_img
process_output(sr_img)
scale = model.scale
padded_size_scaled = tuple(np.multiply(p_shape[0:2],
scale)) + (3, )
scaled_image_shape = tuple(
np.multiply(input_image_array.shape[0:2], scale)) + (3, )
sr_img = stich_together(
collect,
padded_image_shape=padded_size_scaled,
target_shape=scaled_image_shape,
padding_size=padding_size * scale,
)
else:
lr_img = process_array(input_image_array)
sr_img = model.predict(lr_img)[0]
sr_img = process_output(sr_img)
return sr_img