print("Image ordering: {}".format(keras.backend.image_dim_ordering()))
use_N4Correction = True
print("Using N4 correction: {}".format(use_N4Correction))
batch_size = 128
patch_size=(64, 64)
label_size=(8,8)
model = load_model(model_path)
for index, dir in enumerate(listdir(images_dir_path)):
if "brats" in dir:
image_dir_path = os.path.join(images_dir_path, dir)
print("Segmenting image {}".format(image_dir_path))
image, image_dimension = loadTestImage(image_dir_path, use_N4Correction = use_N4Correction)
image = normalize_scans([image], num_channels = 4)[0]
print("Image dimension", image_dimension)
segmentation = segment(image, image_dimension, patch_size, label_size, model.predict, batch_size = batch_size, tf_ordering=tf_ordering)
#save segmentation
segmentation = sitk.GetImageFromArray(segmentation)
segmentation = sitk.Cast(segmentation, sitk.sitkUInt8)
output_file = os.path.join(output_dir, dir) + ".mha"
sitk.WriteImage(segmentation, output_file)
print("Saved image to {}".format(output_file))
print("--- {} seconds ---".format(time.time() - start_time))
shape = (num_channels, patch_size[0], patch_size[1])
if model_file is None:
print("Creating new model")
model = createModel(shape, tf_ordering)
else:
print("Loading model from", model_file)
model = load_model(model_file)
model.summary()
print("Trainable weights", model.trainable_weights)
(images, labels,
image_dimensions) = loadImages(data_dir, use_N4Correction=use_N4Correction)
print("Normalizing scans")
images = normalize_scans(images, 4)
#(val_images, val_labels, val_dimensions) = loadImages(validation_dir, use_N4Correction = use_N4Correction)
assert (image_dimensions == [image.shape for image in images])
#assert(val_dimensions == [image.shape for image in val_images])
print("Loaded %d training images" % len(images))
#print("Loaded %d validation images"%len(val_images))
dataExtractor = DataExtractor(images,
labels, [], [],
tf_ordering=tf_ordering,
patch_size=patch_size,
label_size=label_size)
print("Batch size", batch_size)
shape = (4, patch_size[0], patch_size[1])
if model_file is None:
print("Creating new model")
model = createModel(shape, tf_ordering)
else:
print("Loading model from", model_file)
model = load_model(model_file, custom_objects={'dice': dice})
model.summary()
print("Trainable weights", model.trainable_weights)
(images, labels,
image_dimensions) = loadImages(data_dir, use_N4Correction=use_N4Correction)
print("Normalizing each scan")
images = normalize_scans(images, num_channels=4)
#(val_images, val_labels, val_dimensions) = loadImages(validation_dir, use_N4Correction = use_N4Correction)
assert (image_dimensions == [image.shape for image in images])
print("Loaded %d training images" % len(images))
dataExtractor = DataExtractor(images,
labels,
tf_ordering=tf_ordering,
patch_size=patch_size)
samples_weights = [1, 1, 1, 1, 1]
print("Using weights for data", samples_weights)