def convert_sthv1_csv_to_json(class_file_path, train_csv_path, val_csv_path,
test_csv_path, video_dir_path, dst_json_path):
labels = load_labels(class_file_path)
train_database = convert_csv_to_dict(train_csv_path, 'training')
val_database = convert_csv_to_dict(val_csv_path, 'validation')
if test_csv_path.exists():
test_database = convert_csv_to_dict(test_csv_path, 'testing')
dst_data = {}
dst_data['labels'] = labels
dst_data['database'] = {}
dst_data['database'].update(train_database)
dst_data['database'].update(val_database)
if test_csv_path.exists():
dst_data['database'].update(test_database)
count = 0
for k, v in tqdm.tqdm(dst_data['database'].items()):
if 'label' in v['annotations']:
label = v['annotations']['label']
else:
label = 'test'
video_path = video_dir_path / k
n_frames = get_n_frames(video_path)
v['annotations']['segment'] = (1, n_frames + 1)
v['video_path'] = str(video_path)
# count += 1
# if count == 1000:
# break
with dst_json_path.open('w') as dst_file:
json.dump(dst_data, dst_file)
def convert_photosensitivity_csv_to_json(label_csv_path, train_csv_path, val_csv_path, video_dir_path, dst_json_path):
labels = load_labels(label_csv_path)
train_database = convert_csv_to_dict(train_csv_path, 'training')
val_database = convert_csv_to_dict(val_csv_path, 'validation')
dst_data = {}
dst_data['labels'] = labels
dst_data['database'] = {}
dst_data['database'].update(train_database)
dst_data['database'].update(val_database)
for k, v in dst_data['database'].items():
if v['annotations'] is not None:
label = v['annotations']['label']
else:
label = 'test'
video_path = video_dir_path / label / k
# print(video_path)
n_frames = get_n_frames(video_path)
# print(n_frames)
v['annotations']['segment'] = (1, n_frames + 1)
with dst_json_path.open('w') as dst_file:
json.dump(dst_data, dst_file)
def convert_kinetics_csv_to_json(train_csv_path, val_csv_path, test_csv_path,
video_dir_path, video_type, dst_json_path):
labels = load_labels(train_csv_path)
train_database = convert_csv_to_dict(train_csv_path, 'training')
val_database = convert_csv_to_dict(val_csv_path, 'validation')
if test_csv_path.exists():
test_database = convert_csv_to_dict(test_csv_path, 'testing')
dst_data = {}
dst_data['labels'] = labels
dst_data['database'] = {}
dst_data['database'].update(train_database)
dst_data['database'].update(val_database)
if test_csv_path.exists():
dst_data['database'].update(test_database)
for k, v in dst_data['database'].items():
if 'label' in v['annotations']:
label = v['annotations']['label']
else:
label = 'test'
if video_type == 'jpg':
video_path = video_dir_path / label / k
if video_path.exists():
n_frames = get_n_frames(video_path)
v['annotations']['segment'] = (1, n_frames + 1)
else:
video_path = video_dir_path / label / f'{k}.hdf5'
if video_path.exists():
n_frames = get_n_frames_hdf5(video_path)
v['annotations']['segment'] = (0, n_frames)
with dst_json_path.open('w') as dst_file:
json.dump(dst_data, dst_file)
def convert_ucf101_csv_to_json(label_csv_path, train_csv_path, val_csv_path,
video_dir_path, audio_dir_path, dst_json_path,
video_type):
if audio_dir_path is not None:
labels = [x.name for x in sorted(audio_dir_path.iterdir())]
else:
labels = load_labels(label_csv_path)
train_database = convert_csv_to_dict(train_csv_path, 'training', labels)
val_database = convert_csv_to_dict(val_csv_path, 'validation', labels)
dst_data = {}
dst_data['labels'] = labels
dst_data['database'] = {}
dst_data['database'].update(train_database)
dst_data['database'].update(val_database)
for k, v in dst_data['database'].items():
if v['annotations'] is not None:
label = v['annotations']['label']
else:
label = 'test'
video_path = video_dir_path / label / k
if video_type == 'jpg':
n_frames = get_n_frames(video_path)
else:
video_path = str(video_path) + '.hdf5'
n_frames = get_n_frames_hdf5(video_path)
v['annotations']['segment'] = (1, n_frames + 1)
if audio_dir_path is not None: # save the subset
dst_json_path = dst_json_path.parent / dst_json_path.name.replace(
'ucf101', 'ucf51')
with dst_json_path.open('w') as dst_file:
json.dump(dst_data, dst_file)