def generate_training_list():
subject_id_list = os.listdir(args.db_dir)
subject_id_list.sort()
subject_id = {}
for index, subject_dir in enumerate(subject_id_list):
subject_id[subject_dir] = index
final_contents = []
p_bar = tqdm(total=len(subject_id_list))
for sub_dir in subject_id_list:
identity = subject_id[sub_dir]
cur_dir = os.path.join(args.db_dir, sub_dir)
im_files = os.listdir(cur_dir)
items = [[identity, os.path.join(sub_dir, im_file)] for im_file in im_files]
final_contents.extend(items)
p_bar.update()
p_bar.close()
if args.shuffle:
random.shuffle(final_contents)
save_csv(os.path.join(args.db_dir, 'WebFace.csv'), final_contents, header=['SUBJECT_ID', 'FILE', ])
with open(os.path.join(args.db_dir, 'WebFace.lst'), 'w') as f:
csv_writer = csv.writer(f, delimiter='\t')
for i, item in enumerate(final_contents):
csv_writer.writerow([i, item[0], item[1]])
def merge_train_test_sets():
sub_dirs = os.listdir(args.db_dir)
contents = {}
subject_id_list = []
for sub_dir in sub_dirs:
l = os.listdir(os.path.join(args.db_dir, sub_dir))
l.sort()
contents[sub_dir] = l
subject_id_list.extend(l)
subject_id_list.sort()
subject_id = {}
for index, subject_dir in enumerate(subject_id_list):
subject_id[subject_dir] = index
final_contents = []
p_bar = tqdm(total=len(subject_id_list))
for sub_dir, contents in contents.items():
for ss_dir in contents:
identity = subject_id[ss_dir]
cur_dir = os.path.join(args.db_dir, sub_dir, ss_dir)
im_files = os.listdir(cur_dir)
items = [[identity, os.path.join(sub_dir, ss_dir, im_file)] for im_file in im_files]
final_contents.extend(items)
p_bar.update()
p_bar.close()
if args.shuffle:
random.shuffle(final_contents)
save_csv(os.path.join(args.db_dir, 'VGG-FACE2.csv'), final_contents, header=['FILE', 'SUBJECT_ID'])
with open(os.path.join(args.db_dir, 'VGG-FACE2.lst'), 'w') as f:
csv_writer = csv.writer(f, delimiter='\t')
for i, item in enumerate(final_contents):
csv_writer.writerow([i, item[0], item[1]])
print('Done')
def generate_training_list():
subject_id_list = os.listdir(args.db_dir)
subject_id_list.sort()
subject_id = {}
for index, subject_dir in enumerate(subject_id_list):
if os.path.isdir(os.path.join(args.db_dir, subject_dir)):
subject_id[subject_dir] = index
if not os.path.exists(os.path.join(args.out_dir, subject_dir)):
os.makedirs(os.path.join(args.out_dir, subject_dir))
final_contents = []
p_bar = tqdm(total=len(subject_id_list))
for sub_dir in subject_id_list:
identity = subject_id[sub_dir]
cur_dir = os.path.join(args.db_dir, sub_dir)
im_files = os.listdir(cur_dir)
im_files = [im_file for im_file in im_files if 'texture' in im_file]
items = [[identity, os.path.join(sub_dir, im_file)] for im_file in im_files]
final_contents.extend(items)
p_bar.update()
p_bar.close()
with Pool() as p:
final_contents = list(tqdm(p.imap_unordered(crop_image, final_contents), total=len(final_contents)))
final_contents = list(filter(None, final_contents))
if args.shuffle:
random.shuffle(final_contents)
if len(args.out_file) > 0:
save_csv(args.out_file, final_contents, header=['SUBJECT_ID', 'FILE', ])
with open(args.out_file.replace(os.path.splitext(args.out_file)[1], '.lst'), 'w') as f:
csv_writer = csv.writer(f, delimiter='\t')
for i, item in enumerate(final_contents):
csv_writer.writerow([i, item[0], item[1]])
def main():
contents = load_csv(args.list, read_header=True)
# create out dir
for row in contents:
if not os.path.exists(
os.path.join(args.out_dir, os.path.dirname(row['FILE']))):
os.makedirs(
os.path.join(args.out_dir, os.path.dirname(row['FILE'])))
with Pool() as p:
contents_post = list(
tqdm.tqdm(p.imap_unordered(align_image, contents),
total=len(contents)))
contents_post = list(filter(None, contents_post))
if len(args.out_path) > 0:
save_csv(args.out_path, contents_post, header=['FILE'])