def run(self, input_image, tile_index_z):
in_shape = input_image.shape
pad_shape = self.round(in_shape)
original_image = np.zeros(pad_shape, dtype=input_image.dtype)
original_image[:in_shape[0], :in_shape[1]] = input_image
(original_image_num_pixels_y, original_image_num_pixels_x
) = original_image.shape ###### 180624Change
current_image_num_pixels_y = original_image_num_pixels_y
current_image_num_pixels_x = original_image_num_pixels_x
current_tile_data_space_y = self.tile_num_pixels_y
current_tile_data_space_x = self.tile_num_pixels_x
self.tile_index_z = tile_index_z
self.tile_index_w = 0
images_stride = 1 ###
print('Image size (x, y): ', original_image_num_pixels_x,
original_image_num_pixels_y)
while current_image_num_pixels_y > self.tile_num_pixels_y / 2 or current_image_num_pixels_x > self.tile_num_pixels_x / 2:
current_tile_image_path = self.output_tile_image_path + self.tile_path_wz.format(
self.tile_index_w, self.tile_index_z)
m.mkdir_safe(current_tile_image_path)
# current_image = cv2.resize(original_image,( current_image_num_pixels_x, current_image_num_pixels_y ))
current_image = original_image[::images_stride, ::images_stride]
num_tiles_y = int(
math.ceil(
float(current_image_num_pixels_y) /
self.tile_num_pixels_y))
num_tiles_x = int(
math.ceil(
float(current_image_num_pixels_x) /
self.tile_num_pixels_x))
print('Scale: ', images_stride)
print('Number of panels (x, y): ', num_tiles_x, num_tiles_y)
for iy in range(num_tiles_y):
for ix in range(num_tiles_x):
current_tile_image_name = self.output_tile_image_path + self.tile_images_filename_wzyx.format(
self.tile_index_w, self.tile_index_z, iy, ix)
y = iy * self.tile_num_pixels_y
x = ix * self.tile_num_pixels_x
## tile_image = current_image[y:y + self.tile_num_pixels_y, x:x + self.tile_num_pixels_x] ###
## m.save_tif8(tile_image, current_tile_image_name ) ###
tmp = current_image[y:y + self.tile_num_pixels_y,
x:x + self.tile_num_pixels_x]
tile_with_fringe = np.zeros(
(self.tile_num_pixels_y, self.tile_num_pixels_x),
self.images_dtype)
tile_with_fringe[0:tmp.shape[0], 0:tmp.shape[1]] = tmp
m.save_tif8(tile_with_fringe, current_tile_image_name)
#print(current_tile_image_name)
current_image_num_pixels_y = current_image_num_pixels_y / 2
current_image_num_pixels_x = current_image_num_pixels_x / 2
current_tile_data_space_y = current_tile_data_space_y * 2
current_tile_data_space_x = current_tile_data_space_x * 2
self.tile_index_w = self.tile_index_w + 1
images_stride = images_stride * 2
def run(self, input_ids, tile_index_z):
in_shape = input_ids.shape
pad_shape = self.round(in_shape)
original_ids = np.zeros(pad_shape, dtype=input_ids.dtype)
original_ids[:in_shape[0], :in_shape[1]] = input_ids
(original_ids_num_pixels_y, original_ids_num_pixels_x
) = original_ids.shape ######## ###### 180624Change
## Grow regions until there are no boundaries
current_ids_counts = np.bincount(original_ids.ravel())
current_ids_counts_ids = np.nonzero(current_ids_counts)[0]
current_max = np.max(current_ids_counts_ids)
self.tile_index_z = tile_index_z
if self.id_max < current_max:
self.id_max = current_max
self.id_counts.resize(self.id_max + 1)
self.id_counts[current_ids_counts_ids] = self.id_counts[
current_ids_counts_ids] + np.int64(
current_ids_counts[current_ids_counts_ids])
current_ids_num_pixels_y = original_ids_num_pixels_y
current_ids_num_pixels_x = original_ids_num_pixels_x
current_tile_data_space_y = self.tile_num_pixels_y
current_tile_data_space_x = self.tile_num_pixels_x
self.tile_index_w = 0
ids_stride = 1
while current_ids_num_pixels_y > self.tile_num_pixels_y / 2 or current_ids_num_pixels_x > self.tile_num_pixels_x / 2:
current_tile_ids_path = self.output_tile_ids_path + self.tile_path_wz.format(
self.tile_index_w, self.tile_index_z)
m.mkdir_safe(current_tile_ids_path)
current_ids = original_ids[::ids_stride, ::ids_stride]
num_tiles_y = int(
math.ceil(
float(current_ids_num_pixels_y) / self.tile_num_pixels_y))
num_tiles_x = int(
math.ceil(
float(current_ids_num_pixels_x) / self.tile_num_pixels_x))
print('Scale: ', ids_stride)
print('Number of panels (x, y): ', num_tiles_x, num_tiles_y)
for iy in range(num_tiles_y):
for ix in range(num_tiles_x):
y = iy * self.tile_num_pixels_y
x = ix * self.tile_num_pixels_x
current_tile_ids_name = self.output_tile_ids_path + self.tile_ids_filename_wzyx.format(
self.tile_index_w, self.tile_index_z, iy, ix)
tile_ids = np.zeros(
(self.tile_num_pixels_y, self.tile_num_pixels_x),
np.uint32)
tile_ids_non_padded = current_ids[y:y +
self.tile_num_pixels_y,
x:x +
self.tile_num_pixels_x]
tile_ids[0:tile_ids_non_padded.shape[0],
0:tile_ids_non_padded.
shape[1]] = tile_ids_non_padded[:, :]
if os.path.isfile(
current_tile_ids_name): ## Backup for undo
shutil.move(current_tile_ids_name,
current_tile_ids_name + '_')
self.u_info.ids_files_undo.extend(
current_tile_ids_name)
m.save_hdf5(current_tile_ids_name,
self.u_info.tile_var_name, tile_ids)
for unique_tile_id in np.unique(tile_ids):
self.id_tile_list.append(
(unique_tile_id, self.tile_index_w,
self.tile_index_z, iy, ix))
current_ids_num_pixels_y = current_ids_num_pixels_y / 2
current_ids_num_pixels_x = current_ids_num_pixels_x / 2
current_tile_data_space_y = current_tile_data_space_y * 2
current_tile_data_space_x = current_tile_data_space_x * 2
self.tile_index_w = self.tile_index_w + 1
ids_stride = ids_stride * 2
def _Run(self, parent, params, comm_title):
##
## Remove preovious results.
##
removal_file1 = os.path.join( params['Output Inference Folder'] ,'0','0','seg-0_0_0.npz' )
removal_file2 = os.path.join( params['Output Inference Folder'], '0','0','seg-0_0_0.prob')
if os.path.isfile(removal_file1) or os.path.isfile(removal_file2) :
Reply = QMessageBox.question(parent, 'FFN', 'seg-0_0_0 files were found at the Output Inference Folder. Remove them?', QMessageBox.Yes | QMessageBox.No, QMessageBox.No)
if Reply == QMessageBox.Yes:
with contextlib.suppress(FileNotFoundError):
os.remove(removal_file1)
with contextlib.suppress(FileNotFoundError):
os.remove(removal_file2)
print('seg-0_0_0 files were removed.')
else:
print('FFN inference was canceled.')
return
##
## h5 file (target image file) generation.
##
target_image_file_h5 = os.path.join(params['FFN File Folder'], "grayscale_inf.h5")
try:
target_image_files = m.ObtainImageFiles(params['Target Image Folder'])
images = [cv2.imread(i, cv2.IMREAD_GRAYSCALE) for i in target_image_files]
images = np.array(images)
image_z = images.shape[0]
image_y = images.shape[1]
image_x = images.shape[2]
image_mean = np.mean(images).astype(np.int16)
image_std = np.std(images).astype(np.int16)
print('x: {}, y: {}, z: {}'.format(image_x, image_y, image_z))
with h5py.File( target_image_file_h5 , 'w') as f:
f.create_dataset('raw', data=images, compression='gzip')
print('"grayscale_inf.h5" file (target inference image) was generated.')
except:
print("Error: Target Image h5 was not generated.")
return False
##
## Inference configration file generation
##
request = {}
request['image'] = {"hdf5": "{}@raw".format(target_image_file_h5).replace('\\', '/') }
request['image_mean'] = image_mean
request['image_stddev'] = image_std
request['checkpoint_interval'] = int(params['Checkpoint Interval'])
request['seed_policy'] = "PolicyPeaks"
request['model_checkpoint_path'] = params['Tensorflow Model Files'].replace('\\', '/')
request['model_name'] = "convstack_3d.ConvStack3DFFNModel"
if params['Sparse Z'] != Qt.Unchecked:
request['model_args'] = "{\\\"depth\\\": 9, \\\"fov_size\\\": [33, 33, 17], \\\"deltas\\\": [8, 8, 4]}"
#request['model_args'] = ' {"depth":9,"fov_size":[33,33,17],"deltas":[8,8,4]} '
else :
request['model_args'] = "{\\\"depth\\\": 12, \\\"fov_size\\\": [33, 33, 33], \\\"deltas\\\": [8, 8, 8]}"
#request['model_args'] = ' {"depth":12,"fov_size":[33,33,33],"deltas":[8,8,8]} '
request['segmentation_output_dir'] = params['Output Inference Folder'].replace('\\', '/')
inference_options = {}
inference_options['init_activation'] = 0.95
inference_options['pad_value'] = 0.05
inference_options['move_threshold'] = 0.9
inference_options['min_boundary_dist'] = {"x": 1, "y": 1, "z": 1}
inference_options['segment_threshold'] = 0.6
inference_options['min_segment_size'] = 1000
request['inference_options'] = inference_options
config_file = os.path.join(params['FFN File Folder'], "inference_params.pbtxt")
with open(config_file, "w", encoding='utf-8') as f:
self.write_call(f, request, "")
print('Configuration file was saved at :')
print(config_file)
print('\n')
##
## Inference start (I gave up the use of run_inference because of the augment parsing problem)
##
m.mkdir_safe(os.path.join( params['Output Inference Folder'] ,'0','0' ) )
##
comm_inference = parent.u_info.exec_run_inference
comm_inference = comm_inference.split(' ')
params = ['--image_size_x', np.str( image_x ),
'--image_size_y', np.str( image_y ),
'--image_size_z', np.str( image_z ),
'--parameter_file', config_file
]
comm_inference += params
print(comm_title)
# print(comm_inference)
print('\n')
s.call(comm_inference)
print(comm_title, 'was finished.')
return True
