def get_results_dicts(out_dir, views, image_pairs_dict, n_classes, _continue):
from MultiPlanarUNet.logging import init_result_dicts, save_all, load_result_dicts
if _continue:
csv_dir = os.path.join(out_dir, "csv")
results, detailed_res = load_result_dicts(csv_dir=csv_dir, views=views)
else:
# Prepare dictionary to store results in pd df
results, detailed_res = init_result_dicts(views, image_pairs_dict, n_classes)
# Save to check correct format
save_all(results, detailed_res, out_dir)
return results, detailed_res
def _merged_eval(image_id, pred, true, n_classes, results, per_view_results,
out_dir):
# Calculate combined prediction dice
dices = evaluate(pred, true, n_classes, ignore_zero=True)
mean_dice = dices[~np.isnan(dices)].mean()
per_view_results["MJ"][image_id] = dices
print("Combined dices: ", dices)
print("Combined mean dice: ", mean_dice)
results.loc[image_id, "MJ"] = mean_dice
# Overwrite with so-far results
save_all(results, per_view_results, out_dir)
def _per_view_evaluation(image_id, pred, true, mapped_pred, mapped_true, view,
n_classes, results, per_view_results, out_dir, args):
if np.random.rand() > args.eval_prob:
print("Skipping evaluation for view %s... "
"(eval_prob=%.3f)" % (view, args.eval_prob))
return
# Evaluate the raw view performance
view_dices = evaluate(pred, true, n_classes)
mapped_dices = evaluate(mapped_pred, mapped_true, n_classes)
mean_dice = mapped_dices[~np.isnan(mapped_dices)][1:].mean()
# Print dice scores
print("View dice scores: ", view_dices)
print("Mapped dice scores: ", mapped_dices)
print("Mean dice (n=%i): " % (len(mapped_dices) - 1), mean_dice)
# Add to results
results.loc[image_id, str(view)] = mean_dice
per_view_results[str(view)][image_id] = mapped_dices[1:]
# Overwrite with so-far results
save_all(results, per_view_results, out_dir)
def entry_func(args=None):
# Get command line arguments
args = get_argparser().parse_args(args)
assert_args(args)
# Get most important paths
project_dir = os.path.abspath(args.project_dir)
out_dir = os.path.abspath(args.out_dir)
nii_res_dir = os.path.join(out_dir, "nii_files")
create_folders(nii_res_dir, create_deep=True)
# Check if valid dir structures
validate_folders(project_dir,
out_dir,
overwrite=args.overwrite,
_continue=vars(args)["continue"])
# Get settings from YAML file
hparams = load_hparams(project_dir)
# Get dataset
image_pair_loader, image_pair_dict = get_image_pair_loader(
args, hparams, out_dir)
# Wait for PID to terminate before continuing, if specified
if args.wait_for:
await_PIDs(args.wait_for, check_every=120)
# Set GPU device
num_GPUs = set_gpu_vis(args)
# Get views
views = np.load("%s/views.npz" % project_dir)["arr_0"]
del hparams['fit']['views']
# Prepare result dicts
results, per_view_results = None, None
if not args.no_eval:
results, per_view_results = get_results_dicts(
out_dir, views, image_pair_dict, hparams["build"]["n_classes"],
vars(args)["continue"])
# Get model and load weights, assign to one or more GPUs
model, weights_name = get_model(project_dir, num_GPUs, hparams['build'])
fusion_model = None
if not args.sum_fusion:
fusion_model = get_fusion_model(
n_views=len(views),
n_classes=hparams["build"]["n_classes"],
project_dir=project_dir,
weights_name=weights_name,
num_GPUs=num_GPUs)
run_predictions_and_eval(image_pair_loader=image_pair_loader,
image_pair_dict=image_pair_dict,
model=model,
fusion_model=fusion_model,
views=views,
hparams=hparams,
args=args,
results=results,
per_view_results=per_view_results,
out_dir=out_dir,
nii_res_dir=nii_res_dir)
if not args.no_eval:
# Write final results
save_all(results, per_view_results, out_dir)