def get_space_connected_to_position(layer, position):
empty_mask = (layer != Tiles.wall)
labeled_components, component_count = label_connected_components(
empty_mask)
label = labeled_components[tuple(position)]
connected_space = (labeled_components == label)
return connected_space
def get_random_positions_for_lock(layer, start_position):
connected_space = MissionGenerator.get_space_connected_to_position(
layer, start_position)
connected_walls_and_corridors = MissionGenerator.get_walls_and_corridors_connected_to_space(
layer, connected_space)
connected_walls_and_corridors_components, component_count = label_connected_components(
connected_walls_and_corridors)
random_positions = MissionGenerator.get_random_positions_per_component(
connected_walls_and_corridors_components, component_count, 1)
return random_positions
def get_space_component_masks(layer):
masks = []
empty_space = (layer == Tiles.empty)
connected_components, component_count = label_connected_components(
empty_space)
for component_label in range(1, component_count + 1):
component_mask = (
connected_components == component_label).astype(int)
masks.append(component_mask)
return masks
def remove_hazard_component(layer, hazard_position, hazard_tile):
hazard_mask = (layer == hazard_tile)
labeled_components, component_count = label_connected_components(
hazard_mask)
hazard_label = labeled_components[hazard_position]
layer[labeled_components == hazard_label] = Tiles.empty
sub = sitk.ReadImage('./segmentation-0.nii')
# Get Numpy data and compress to int8.
reference_volume = sitk.GetArrayFromImage(ref)#.astype(np.int8)
lreference_volume = reference_volume.view((1,75,512,512))
submission_volume = sitk.GetArrayFromImage(sub).astype(np.int8)
# Ensure that the shapes of the masks match.
if submission_volume.shape!=reference_volume.shape:
raise AttributeError("Shapes do not match! Prediction mask {}, "
"ground truth mask {}"
"".format(submission_volume.shape,
reference_volume.shape))
# Create masks with labeled connected components.
# (Assuming there is always exactly one liver - one connected comp.)
pred_mask_liver = submission_volume;pred_mask_liver[submission_volume>=1]=1
pred_mask_lesion, num_predicted = label_connected_components(submission_volume==2, output=np.int16)#default structuring is 2-D array, so the input demension is?
true_mask_liver = reference_volume;true_mask_liver[reference_volume>=1]=1
true_mask_lesion, num_reference = label_connected_components(reference_volume==2, output=np.int16)
liver_prediction_exists = np.any(submission_volume==1)
# Compute per-case (per patient volume) dice.
if not np.any(pred_mask_lesion) and not np.any(true_mask_lesion):
tumor_dice_per_case = 1.
else:
tumor_dice_per_case = metric.dc(pred_mask_lesion, true_mask_lesion)
if liver_prediction_exists:
liver_dice_per_case = metric.dc(pred_mask_liver, true_mask_liver)
else:
liver_dice_per_case = 0
print(tumor_dice_per_case)
# Get Numpy data and compress to int8.
reference_volume = (reference_volume.get_data()).astype(np.int8)
submission_volume = (submission_volume.get_data()).astype(np.int8)
# Ensure that the shapes of the masks match.
if submission_volume.shape != reference_volume.shape:
raise AttributeError("Shapes do not match! Prediction mask {}, "
"ground truth mask {}"
"".format(submission_volume.shape,
reference_volume.shape))
print("Done loading files ({:.2f} seconds)".format(t()))
# Create lesion and liver masks with labeled connected components.
# (Assuming there is always exactly one liver - one connected comp.)
pred_mask_lesion, num_predicted = label_connected_components( \
submission_volume==2, output=np.int16)
true_mask_lesion, num_reference = label_connected_components( \
reference_volume==2, output=np.int16)
pred_mask_liver = submission_volume >= 1
true_mask_liver = reference_volume >= 1
liver_prediction_exists = np.any(submission_volume == 1)
print("Done finding connected components ({:.2f} seconds)".format(t()))
# Identify detected lesions.
# Retain detected_mask_lesion for overlap > 0.5
for overlap in [0, 0.5]:
detected_mask_lesion, mod_ref_mask, num_detected = detect_lesions( \
prediction_mask=pred_mask_lesion,
reference_mask=true_mask_lesion,
min_overlap=overlap)
def fill_unused_space(level):
start_position = np.argwhere(level.upper_layer == Tiles.player)[0]
non_wall_mask = (level.upper_layer != Tiles.wall).astype(int)
connected_components, component_count = label_connected_components(non_wall_mask)
used_space_component = connected_components[tuple(start_position)]
level.upper_layer[connected_components != used_space_component] = Tiles.wall
def get_potential_lock_components(layer):
potential_lock_mask = MissionGenerator.get_wall_corridor_mask(layer)
labeled_components, component_count = label_connected_components(
potential_lock_mask)
return labeled_components, component_count
def get_rooms_components(layer):
empty_mask = (layer == Tiles.empty)
labeled_components, component_count = label_connected_components(
empty_mask)
return labeled_components, component_count