def other_instance_image(self, target_label, other_labels):
"""
Returns the mask, of neighboring other_labels for a given target_label.
:param target_label: Image, where pixels of the target label are set to 1, all other pixels are 0.
:param other_labels: List of images of all labels. Pixels of the label are set to 1, all other pixels are 0.
:return: The image (np array), where pixels of neighboring labels are set to 2, all other pixels are 0.
"""
channel_axis = self.get_channel_axis(target_label, self.data_format)
dim = len(target_label.shape)
mask = np.zeros_like(target_label)
circle_radius = self.image_size[0] * self.instance_image_radius_factor
# handle images with dim 3 (videos) differently
if dim == 3:
com = center_of_mass(target_label)
com = com[1:]
for i in range(target_label.shape[channel_axis]):
current_slice = [slice(None), slice(None)]
current_slice.insert(channel_axis, slice(i, i + 1))
current_mask = np.squeeze(mask[tuple(current_slice)],
axis=channel_axis)
draw_circle(current_mask, com, circle_radius)
else:
com = center_of_mass(target_label)
draw_circle(mask, com, circle_radius)
other_instances = np.zeros_like(mask)
for other_label in other_labels:
if np.any(np.bitwise_and(mask == 1, other_label == 1)):
other_instances += other_label
other_instances[other_instances > 0] = 2
return other_instances
def other_instance_image(self, target_label, other_labels):
"""
Returns the mask, of neighboring other_labels for a given target_label.
:param target_label: Image, where pixels of the target label are set to 1, all other pixels are 0.
:param other_labels: List of images of all labels. Pixels of the label are set to 1, all other pixels are 0.
:return: The image (np array), where pixels of neighboring labels are set to 2, all other pixels are 0.
"""
frame_axis = 0
dim = len(target_label.shape)
mask = np.copy(target_label) #np.zeros_like(target_label)
if self.loss_mask_dilation_size > 0:
if dim == 3:
for i in range(target_label.shape[frame_axis]):
current_slice = [slice(None), slice(None)]
current_slice.insert(frame_axis, slice(i, i + 1))
mask[tuple(current_slice)] = dilation_circle(
np.squeeze(mask[tuple(current_slice)]),
(self.loss_mask_dilation_size,
self.loss_mask_dilation_size))
else:
mask = dilation_circle(mask, (self.loss_mask_dilation_size,
self.loss_mask_dilation_size))
circle_radius = self.image_size[
0] * self.instance_image_radius_factor / self.scale_factor[0]
# handle images with dim 3 (videos) differently
if dim == 3:
com = center_of_mass(target_label)
com = com[1:]
is_in_frame = np.any(np.any(target_label, axis=1), axis=1)
min_index = np.maximum(np.min(np.where(is_in_frame)) - 1, 0)
max_index = np.minimum(
np.max(np.where(is_in_frame)) + 1,
target_label.shape[frame_axis])
#for i in range(target_label.shape[frame_axis]):
for i in range(min_index, max_index):
current_slice = [slice(None), slice(None)]
current_slice.insert(frame_axis, slice(i, i + 1))
#com = center_of_mass(np.squeeze(target_label[tuple(current_slice)]))
current_mask = np.squeeze(mask[tuple(current_slice)],
axis=frame_axis)
draw_circle(current_mask, com, circle_radius)
else:
com = center_of_mass(target_label)
draw_circle(mask, com, circle_radius)
other_instances = np.zeros_like(mask)
for other_label in other_labels:
if np.array_equal(target_label, other_label):
continue
if np.any(np.bitwise_and(mask == 1, other_label == 1)):
other_instances += other_label
other_instances[other_instances > 0] = 1
return other_instances
def visualize_landmark(self, image_canvas, landmark, color, annotation, annotation_color):
"""
Visualize a single landmark.
:param image_canvas: The image canvas object.
:param landmark: The landmark.
:param color: The landmark color.
:param annotation: The annotation string.
:param annotation_color: The annotation color.
"""
coords = landmark.coords
if self.spacing is not None:
coords = coords / self.spacing
draw_circle(image_canvas, [coords[1], coords[0]], self.radius, color)
def current_instance_neighbors(target_label, label_image, image_size, instance_image_radius_factor, loss_mask_dilation_size):
"""
Returns the mask, of neighboring other_labels for a given target_label.
:param target_label: Image, where pixels of the target label are set to 1, all other pixels are 0.
:param other_labels: List of images of all labels. Pixels of the label are set to 1, all other pixels are 0.
:return: The image (np array), where pixels of neighboring labels are set to 2, all other pixels are 0.
"""
frame_axis = 0
dim = len(label_image.shape)
mask = (label_image == target_label).astype(np.uint8)
if loss_mask_dilation_size > 0:
if dim == 3:
for i in range(label_image.shape[frame_axis]):
current_slice = [slice(None), slice(None)]
current_slice.insert(frame_axis, slice(i, i + 1))
mask[tuple(current_slice)] = dilation_circle(np.squeeze(mask[tuple(current_slice)]), (loss_mask_dilation_size, loss_mask_dilation_size))
else:
mask = dilation_circle(mask, (loss_mask_dilation_size, loss_mask_dilation_size))
circle_radius = image_size[0] * instance_image_radius_factor
# handle images with dim 3 (videos) differently
if dim == 3:
com = center_of_mass(mask)
com = com[1:]
is_in_frame = np.any(np.any(mask, axis=1), axis=1)
min_index = np.maximum(np.min(np.where(is_in_frame)) - 1, 0)
max_index = np.minimum(np.max(np.where(is_in_frame)) + 2, label_image.shape[frame_axis])
for i in range(min_index, max_index):
current_slice = [slice(None), slice(None)]
current_slice.insert(frame_axis, slice(i, i + 1))
if i == min_index:
next_slice = [slice(None), slice(None)]
next_slice.insert(frame_axis, slice(i + 1, i + 2))
mask[tuple(current_slice)] = np.logical_or(mask[tuple(next_slice)], mask[tuple(current_slice)])
if i == max_index - 1:
previous_slice = [slice(None), slice(None)]
previous_slice.insert(frame_axis, slice(i - 1, i))
mask[tuple(current_slice)] = np.logical_or(mask[tuple(previous_slice)], mask[tuple(current_slice)])
current_mask = np.squeeze(mask[tuple(current_slice)], axis=frame_axis)
draw_circle(current_mask, com, circle_radius)
else:
com = center_of_mass(label_image)
draw_circle(mask, com, circle_radius)
current_neighbors = [int(x) for x in np.unique(label_image * mask) if x != 0 and x != target_label]
return current_neighbors