def main():
global config
args = get_args.train25d()
config = path_utils.update_is_augment(args, config)
data_path, _, _, _, _ = path_utils.get_training_h5_paths(BRATS_DIR, args)
if args.overwrite or not os.path.exists(data_path):
prepare_data(args)
train(args)
def main():
args = get_args.train25d()
depth_unet = args.depth_unet
n_base_filters_unet = args.n_base_filters_unet
patch_shape = args.patch_shape
is_crf = args.is_crf
batch_size = args.batch_size
is_hist_match = args.is_hist_match
loss = args.loss
header = ("dice_WholeTumor", "dice_TumorCore", "dice_EnhancingTumor")
model_scores = list()
model_ids = list()
for is_augment in ["1"]:
args.is_augment = is_augment
for model_name in ["unet"]:
args.model = model_name
for is_denoise in ["0"]:
args.is_denoise = is_denoise
for is_normalize in ["z"]:
args.is_normalize = is_normalize
for is_hist_match in ["0"]:
args.is_hist_match = is_hist_match
for loss in ["weighted"]:
args.loss = loss
# for patch_shape in ["160-192-3", "160-192-5", "160-192-7", "160-192-9", "160-192-11", "160-192-13", "160-192-15", "160-192-17"]:
for patch_shape in [
"160-192-11", "160-192-13", "160-192-15",
"160-192-17"
]:
args.patch_shape = patch_shape
args.model_dim = 25
print("=" * 120)
print(
">> processing model-{}{}, depth-{}, filters-{}, patch_shape-{}, is_denoise-{}, is_normalize-{}, is_hist_match-{}, loss-{}"
.format(model_name, args.model_dim,
depth_unet, n_base_filters_unet,
patch_shape, is_denoise,
is_normalize, is_hist_match, loss))
is_test = "0"
predict(args)
# print("="*60)
print(">> finished")
print("=" * 120)
gc.collect()
from keras import backend as K
K.clear_session()
print(list_already_predicted)
print(len(list_already_predicted))
model_path = os.path.join(
BRATS_DIR, "database/model", model_filename)
if os.path.exists(model_path):
print("{} exists. Will skip!!".format(model_path))
else:
try:
print(">> RUNNING:", cmd)
except:
print("something wrong")
from keras import backend as K
os.system(cmd)
K.clear_session()
args = get_args.train25d()
task = "brats/train25d"
args.is_test = "0"
model_list = list()
cmd_list = list()
out_file_list = list()
for is_augment in ["1"]:
args.is_augment = is_augment
for model_name in ["unet"]:
args.model = model_name
for is_denoise in ["0"]:
args.is_denoise = is_denoise
for is_normalize in ["z"]:
args.is_normalize = is_normalize
def main():
args = get_args.train25d()
depth_unet = args.depth_unet
n_base_filters_unet = args.n_base_filters_unet
patch_shape = args.patch_shape
is_crf = args.is_crf
batch_size = args.batch_size
is_hist_match = args.is_hist_match
loss = args.loss
header = ("dice_WholeTumor", "dice_TumorCore", "dice_EnhancingTumor")
model_scores = list()
model_ids = list()
for is_augment in ["1"]:
args.is_augment = is_augment
for model_name in ["unet"]:
args.model = model_name
for is_denoise in ["0"]:
args.is_denoise = is_denoise
for is_normalize in ["z"]:
args.is_normalize = is_normalize
for is_hist_match in ["0"]:
args.is_hist_match = is_hist_match
for loss in ["weighted"]:
args.loss = loss
for patch_shape in ["160-192-3", "160-192-5", "160-192-7", "160-192-9", "160-192-11", "160-192-13", "160-192-15", "160-192-17"]:
# for patch_shape in ["160-192-3", "160-192-5", "160-192-7", "160-192-9", "160-192-11"]:
# for patch_shape in ["160-192-3", "160-192-13", "160-192-15", "160-192-17"]:
args.patch_shape = patch_shape
model_dim = 25
print("="*120)
print(
">> processing model-{}{}, depth-{}, filters-{}, patch_shape-{}, is_denoise-{}, is_normalize-{}, is_hist_match-{}, loss-{}".format(
model_name,
model_dim,
depth_unet,
n_base_filters_unet,
patch_shape,
is_denoise,
is_normalize,
is_hist_match,
loss))
is_test = "0"
model_score, model_path = evaluate(args)
if model_score is not None:
print("="*120)
print(">> finished:")
model_ids.append(
get_filename_without_extension(model_path))
row = get_model_info_header(args.challenge, args.year, args.image_shape, args.is_bias_correction,
args.is_denoise, args.is_normalize, args.is_hist_match,
model_name, model_dim, patch_shape, loss,
depth_unet, n_base_filters_unet)
score = [np.mean(model_score["dice_WholeTumor"]),
np.mean(
model_score["dice_TumorCore"]),
np.mean(
model_score["dice_EnhancingTumor"]),
(np.mean(model_score["dice_WholeTumor"])+np.mean(model_score["dice_TumorCore"])+np.mean(model_score["dice_EnhancingTumor"]))/3]
row.extend(score)
model_scores.append(row)
header = ("challenge", "year", "image_shape", "is_bias_correction",
"is_denoise", "is_normalize", "is_hist_match",
"model_name", "model_dim", "depth_unet",
"n_base_filters_unet", "loss", "patch_shape",
"dice_WholeTumor", "dice_TumorCore",
"dice_EnhancingTumor", "dice_Mean")
final_df = pd.DataFrame.from_records(
model_scores, columns=header, index=model_ids)
print(final_df)
prediction_df_csv_folder = os.path.join(
BRATS_DIR, "database/prediction/csv/")
make_dir(prediction_df_csv_folder)
to_file = prediction_df_csv_folder + "compile.csv"
final_df.to_csv(to_file)