M, C_pc, C_rpc, p, q,
loss_fun='square_loss', alpha=0.2)
H = - np.sum(dP * np.log(dP + 1e-5))
# print(k, H)
P = P + dP
# print(k, np.argmax(dP, axis=1))
re_ranking[k] = np.argmax(P, axis=1).tolist()
return re_ranking
if __name__ == '__main__':
golden_annotations = defaultdict(list)
for case in golden_cases:
file = os.path.join(golden_train_dir, case, "annotation.txt")
annotation = DataReader.parse_annotation(file)
for k, v in annotation.items():
golden_annotations[k].append(np.asarray(v))
t = tqdm(train_cases)
for case in t:
raw_annotation = os.path.join(train_dir, case, "annotation.txt")
annotation = {k: np.asarray(v)
for k, v in DataReader.parse_annotation(raw_annotation).items()}
re_ranking = align_annotation(
golden_annotations, annotation, align_method='FGW')
if re_ranking:
re_annotation = {k: np.asarray([v[i] for i in re_ranking[k]])
for k, v in annotation.items()}
save_refined_annotation(sorted_train_dir, case, re_annotation)
else:
fail_list.write(case + "\n")
results = {}
for color in point_clouds:
results[color] = []
for x, y in point_clouds[color]:
if x > 300 and y > 300:
results[color].append([x, y])
return results
if __name__ == "__main__":
target_case = os.listdir(target_folder)
failure_count = 0
cannot_fix = 0
with tqdm(target_case) as t:
for case_name in t:
point_clouds = DataReader.parse_annotation(
os.path.join(target_folder, case_name, file_name))
stats = {c: len(point_clouds[c]) for c in point_clouds}
if not landmark_num_checker(point_clouds):
print("case '{}' is not legal\n{}".format(
case_name, json.dumps(point_clouds)))
new_point_clouds = location_filter(point_clouds)
if not landmark_num_checker(new_point_clouds):
print(
"location filtered case '{}' is not legal\n{}".format(
case_name, json.dumps(new_point_clouds)))
cannot_fix += 1
failure_count += 1
t.set_postfix(stats)
print("Summary")
print("In folder {}, {} failures found, {} cannot be auto-fixed".format(
target_folder, failure_count, cannot_fix))