def create_weights_from_masks(mask_folder, output_folder, NUM_PROCS=8):
"""Creates weight image for all masks in mask_folder and saves to output folder.
This process uses paralell processing to speed this up.
Parameters
----------
mask_folder : str
Path to mask folder
output_folder : str
Desired output folder
NUM_PROCS : int, optional
How many threads to spawn
"""
from multiprocessing import Pool
if not os.path.exists(output_folder):
os.makedirs(output_folder)
masks = get_all_files(mask_folder)
input_pool = []
for iMask in masks:
image_path = os.path.basename(iMask)
output_path = os.path.join(output_folder, image_path)
input_pool.append([iMask, output_path])
pool = Pool(processes=NUM_PROCS)
pool.map(process_one, input_pool)
def _get_image_paths(self, path):
return fileops.get_all_files(path)
if __name__ == "__main__":
# TODO: Add argparse
"""
image_test = '/home/adamf/data/Karolinska/images/Nuclei_3.png'
mq = MakeQuery('karolinska-188312', 'karolinska', 'v3')
response = mq.get_prediction([image_test])
mask_response = response['predictions'][0]['classes']
org_image, _ = load_image(image_test, size=(512, 512))
def apply_tinge(original, prediction):
# Load and make the predicted cells a bit more red
original = np.stack([original.copy()] * 3, 2)
x_idx, y_idx = np.where(prediction)
original[x_idx, y_idx, 1] /= 5
original[x_idx, y_idx, 2] /= 5
plot_image_from_array(original, show_type='matplotlib')
# Simple argmax prediction
apply_tinge(org_image, mask_response)
# More strict predictions
mask_probability = np.array(response['predictions'][0]['probabilities'])
for i_level in [0.25, 0.5, 0.75]:
apply_tinge(org_image, mask_probability[:, :, 1] > i_level)
"""
all_images = get_all_files("/home/adamf/data/Karolinska/images/")
mq = MakeQuery('karolinska-188312', 'karolinska', 'v3')
n_images = len(all_images)
for i, image in enumerate(all_images):
response = mq.get_prediction([image])
'--output_location',
help='Path to save the output prediction',
required=True
)
parser.add_argument(
'--energy_threshold',
help='Which energy contour to plot',
default=0,
type=int
)
args = parser.parse_args()
if not os.path.exists(args.output_location):
os.makedirs(args.output_location)
all_images = get_all_files(args.image_folder)
mask_predictor = get_predictor(args.mask_model)
watershed_predictor = get_predictor(args.watershed_model)
n_images = len(all_images)
for i, image in enumerate(all_images):
# Get a mask prediction
safe = image_to_websafe_mask(image)
mask_response = mask_predictor(safe)
predicted_mask = np.squeeze(mask_response['classes'])
# Pipe to watershed prediction
safe = image_to_websafe_watershed(image, predicted_mask)
mask_response = watershed_predictor(safe)
predicted_energy = np.squeeze(mask_response['classes'])
img = Image.fromarray(apply_tinge(image, predicted_energy, args.energy_threshold))
from src.lib.imgops import create_weight_file
from src.lib.fileops import get_all_files
for ifile in get_all_files("/home/adamf/data/Karolinska/labels"):
create_weight_file(ifile, "/home/adamf/data/Karolinska/weights_10_25")
parser.add_argument(
'--prediction_model',
help='Path to the prediction model (folder with .pb file)',
required=True)
parser.add_argument('--image_folder',
help='Path to folder containing images to segment',
required=True)
parser.add_argument(
'--mask_folder',
help='Path to folder containing segment masks (binary)',
required=True)
parser.add_argument('--output_location',
help='Path to save the output prediction',
required=True)
args = parser.parse_args()
if not os.path.exists(args.output_location):
os.makedirs(args.output_location)
all_images = get_all_files(args.image_folder)
all_masks = get_all_files(args.mask_folder)
predict = get_predictor(args.prediction_model)
n_images = len(all_images)
for i, (image, mask) in enumerate(zip(all_images, all_masks)):
safe = image_to_websafe(image, mask)
response = predict(safe)
predicted_energy = np.squeeze(response['classes'])
img = Image.fromarray(apply_tinge(image, predicted_energy))
save_path = os.path.join(args.output_location, os.path.basename(image))
img.save(save_path)
def _get_image_paths(self, path):
"""Fetches all files in a given path."""
return fileops.get_all_files(path)