def rebuild_load_data():
normal_videos = read_file(
os.path.join(constants.PATH_UCFCRIME2LOCAL_README,
'normal_videos.txt'))
normal_videos_from_anormal = read_file(
os.path.join(constants.PATH_UCFCRIME2LOCAL_README,
'normal_videos_from_anormal.txt'))
anomaly_normal = read_file(
os.path.join(constants.PATH_UCFCRIME2LOCAL_README,
'anomaly_normal.txt'))
violence_videos = read_file(
os.path.join(constants.PATH_UCFCRIME2LOCAL_README,
'violence_videos.txt'))
X, y, numFrames = [], [], []
ll = [violence_videos, anomaly_normal]
llabels = [1, 0]
for i, videos_list in enumerate(ll):
for v_video in videos_list:
if llabels[i] == 1:
v_path = os.path.join(
constants.PATH_UCFCRIME2LOCAL_FRAMES_VIOLENCE, v_video)
else:
v_path = os.path.join(
constants.PATH_UCFCRIME2LOCAL_FRAMES_NONVIOLENCE, v_video)
l_frames = os.listdir(v_path)
X.append(v_video)
y.append(llabels[i])
numFrames.append(len(l_frames))
save_csvfile_multicolumn(
zip(X, y, numFrames),
os.path.join(constants.PATH_UCFCRIME2LOCAL_README, 'All_data.txt'))
return X, y, numFrames
def crime2localgGetSplit(X, y, numFrames, splits=5):
# print(X)
kfold = KFold(splits, shuffle=True)
if not os.path.exists(
os.path.join(constants.PATH_UCFCRIME2LOCAL_README,
'fold-1-train.txt')):
for i, (train_idx, test_idx) in enumerate(kfold.split(X)):
save_file(
train_idx,
os.path.join(constants.PATH_UCFCRIME2LOCAL_README,
'fold-{}-train.txt'.format(i + 1)))
save_file(
test_idx,
os.path.join(constants.PATH_UCFCRIME2LOCAL_README,
'fold-{}-test.txt'.format(i + 1)))
for i in range(splits):
train_idx = read_file(
os.path.join(constants.PATH_UCFCRIME2LOCAL_README,
'fold-{}-train.txt'.format(i + 1)))
test_idx = read_file(
os.path.join(constants.PATH_UCFCRIME2LOCAL_README,
'fold-{}-test.txt'.format(i + 1)))
train_idx = list(map(int, train_idx))
test_idx = list(map(int, test_idx))
yield train_idx, test_idx
def get_Fold_Data(fold):
train_idx = read_file(
os.path.join(constants.PATH_UCFCRIME2LOCAL_README,
'fold-{}-train.txt'.format(fold)))
test_idx = read_file(
os.path.join(constants.PATH_UCFCRIME2LOCAL_README,
'fold-{}-test.txt'.format(fold)))
train_idx = list(map(int, train_idx))
test_idx = list(map(int, test_idx))
return train_idx, test_idx
def load_fold_data(dataset, fold):
if dataset == 'hockey' or dataset == 'ucfcrime2local' or dataset == 'vif':
if dataset == 'vif':
folder = constants.PATH_VIF_README
else:
folder = constants.PATH_UCFCRIME2LOCAL_README if dataset == 'ucfcrime2local' else constants.PATH_HOCKEY_README
train_idx = read_file(
os.path.join(folder, 'fold_{}_train.txt'.format(fold)))
test_idx = read_file(
os.path.join(folder, 'fold_{}_test.txt'.format(fold)))
train_idx = list(map(int, train_idx))
test_idx = list(map(int, test_idx))
return train_idx, test_idx
def hockeyTrainTestSplit(split_type, datasetAll, labelsAll, numFramesAll):
train_idx = read_file(
os.path.join(
constants.PATH_HOCKEY_README,
'fold_{}_train.txt'.format(int(split_type[len(split_type) - 1]))))
test_idx = read_file(
os.path.join(
constants.PATH_HOCKEY_README,
'fold_{}_test.txt'.format(int(split_type[len(split_type) - 1]))))
train_idx = list(map(int, train_idx))
test_idx = list(map(int, test_idx))
train_x = list(itemgetter(*train_idx)(datasetAll))
train_y = list(itemgetter(*train_idx)(labelsAll))
train_numFrames = list(itemgetter(*train_idx)(numFramesAll))
test_x = list(itemgetter(*test_idx)(datasetAll))
test_y = list(itemgetter(*test_idx)(labelsAll))
test_numFrames = list(itemgetter(*test_idx)(numFramesAll))
return train_x, train_y, train_numFrames, test_x, test_y, test_numFrames
def k_folds(n_splits, subjects, splits_folder):
"""
Generates folds for cross validation
Args:
n_splits: folds number
subjects: number of patients
frames: length of the sequence of each patient
"""
indices = np.arange(subjects).astype(int)
if n_splits == 1:
for i in range(1):
train, test = train_test_split(80, 20, subjects)
yield train, test
# indices = np.arange(subjects * frames).astype(int)
else:
if not os.path.exists(os.path.join(splits_folder, 'fold_1_train.txt')):
# if True:
for fold, test_idx in enumerate(get_indices(n_splits, subjects)):
train_idx = np.setdiff1d(indices, test_idx)
save_file(
train_idx,
os.path.join(splits_folder,
'fold_' + str(fold + 1) + '_train.txt'))
save_file(
test_idx,
os.path.join(splits_folder,
'fold_' + str(fold + 1) + '_test.txt'))
yield train_idx, test_idx
else:
for fold in range(n_splits):
train_idx = read_file(
os.path.join(splits_folder,
'fold_' + str(fold + 1) + '_train.txt'))
test_idx = read_file(
os.path.join(splits_folder,
'fold_' + str(fold + 1) + '_test.txt'))
train_idx = list(map(int, train_idx))
test_idx = list(map(int, test_idx))
yield train_idx, test_idx
def plot_bbox_annotations():
anomaly_normal = read_file(
os.path.join(constants.PATH_UCFCRIME2LOCAL_README,
'anomaly_normal.txt'))
for av in anomaly_normal:
v_path = os.path.join(constants.PATH_UCFCRIME2LOCAL_FRAMES_NONVIOLENCE,
av)
print(v_path)
l_frames = os.listdir(v_path)
l_frames.sort(key=lambda f: int("".join(filter(str.isdigit, f))))
av = av[:-8]
bdx_file_path = os.path.join(
constants.PATH_UCFCRIME2LOCAL_Txt_ANNOTATIONS, av + '.txt')
data = []
with open(bdx_file_path, 'r') as file:
for row in file:
data.append(row.split())
# data = np.array(data)
gt_bboxes = []
counter = 0
start = False
end = False
for i, frame_path in enumerate(l_frames):
# print('------frame_path=',frame_path)
pth, frame_name = os.path.split(frame_path)
splits = re.split('(\d+)', frame_name)
frame_number = int(splits[1])
if frame_number >= len(data):
break
frame_data = data[frame_number]
# print('video={}, frame={}, frame_number={}, gt={}'.format(video_name, frame_name, frame_number, frame_data))
if frame_number != int(frame_data[5]):
print('=========*********** Error en Ground Truth!!!!!!!!!')
break
x0, y0, w, h = int(frame_data[1]), int(
frame_data[2]), int(frame_data[3]) - int(frame_data[1]), int(
frame_data[4]) - int(frame_data[2])
gt_bboxes.append([x0, y0, w, h])
frame = cv2.imread(os.path.join(v_path, frame_path))
flac = int(frame_data[6])
if flac == 0:
cv2.rectangle(frame, (x0, y0), (x0 + w, y0 + h), (0, 255, 0),
2)
cv2.imshow("frame", frame)
key = cv2.waitKey(0)
def temporal_cut_long_videos():
anomaly_normal = read_file(
os.path.join(constants.PATH_UCFCRIME2LOCAL_README,
'anomaly_normal.txt'))
for av in anomaly_normal:
v_path = os.path.join(constants.PATH_UCFCRIME2LOCAL_FRAMES_NONVIOLENCE,
av)
print(v_path)
l_frames = os.listdir(v_path)
l_frames.sort(key=lambda f: int("".join(filter(str.isdigit, f))))
bdx_file_path = os.path.join(
constants.PATH_UCFCRIME2LOCAL_Txt_ANNOTATIONS, av + '.txt')
data = []
with open(bdx_file_path, 'r') as file:
for row in file:
data.append(row.split())
# data = np.array(data)
gt_bboxes = []
anomaly_clips = []
counter = 0
start = False
end = False
clip_frames = []
counter = 0
for i, frame_path in enumerate(l_frames):
# print('------frame_path=',frame_path)
pth, frame_name = os.path.split(frame_path)
splits = re.split('(\d+)', frame_name)
frame_number = int(splits[1])
if frame_number >= len(data):
break
frame_data = data[frame_number]
# print('video={}, frame={}, frame_number={}, gt={}'.format(video_name, frame_name, frame_number, frame_data))
if frame_number != int(frame_data[5]):
print('=========*********** Error en Ground Truth!!!!!!!!!')
break
x0, y0, w, h = int(frame_data[1]), int(
frame_data[2]), int(frame_data[3]) - int(frame_data[1]), int(
frame_data[4]) - int(frame_data[2])
gt_bboxes.append([x0, y0, w, h])
# frame = cv2.imread(os.path.join(v_path,frame_path))
flac = int(frame_data[6])
if flac == 0:
# cv2.rectangle(frame, (x0, y0),(x0+w, y0+h), (0,255,0), 2)
clip_frames.append(os.path.join(v_path, frame_path))
if not start:
# print('================')
# clip_frames.append(os.path.join(v_path,frame_path))
start = True
elif start:
end = True
start = False
# for f in clip_frames:
# print(f)
anomaly_clips.append(clip_frames)
counter += 1
f_name = av + '-VSplit-' + str(counter)
if not os.path.isdir(
os.path.join(
constants.PATH_UCFCRIME2LOCAL_FRAMES_NONVIOLENCE,
f_name)):
os.mkdir(
os.path.join(
constants.PATH_UCFCRIME2LOCAL_FRAMES_NONVIOLENCE,
f_name))
for fimg in clip_frames:
newPath = shutil.copy(
fimg,
os.path.join(
constants.PATH_UCFCRIME2LOCAL_FRAMES_NONVIOLENCE,
f_name))
clip_frames = []
def customize_kfold(n_splits, dataset, X, y, shuffle=True):
# def customize_kfold(n_splits, dataset, shuffle=True):
# X=np.arange(X_len)
if dataset == 'hockey' or dataset == 'ucfcrime2local':
kfold = StratifiedKFold(n_splits, shuffle=shuffle)
folder = constants.PATH_UCFCRIME2LOCAL_README if dataset == 'ucfcrime2local' else constants.PATH_HOCKEY_README
if not os.path.exists(os.path.join(folder, 'fold_1_train.txt')):
# X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)
for i, (train_idx, test_idx) in enumerate(kfold.split(X, y)):
save_file(
train_idx,
os.path.join(folder, 'fold_{}_train.txt'.format(i + 1)))
save_file(
test_idx,
os.path.join(folder, 'fold_{}_test.txt'.format(i + 1)))
for i in range(n_splits):
train_idx = read_file(
os.path.join(folder, 'fold_{}_train.txt'.format(i + 1)))
test_idx = read_file(
os.path.join(folder, 'fold_{}_test.txt'.format(i + 1)))
train_idx = list(map(int, train_idx))
test_idx = list(map(int, test_idx))
yield train_idx, test_idx
elif dataset == 'vif':
splitsLen = []
folder = os.path.join(constants.PATH_VIF_README)
if not os.path.exists(os.path.join(folder, 'fold_1_train.txt')):
if not os.path.exists(os.path.join(folder, 'lengths.txt')):
for fold in range(n_splits):
violence_path = os.path.join(constants.PATH_VIF_FRAMES,
str(fold + 1), 'Violence')
non_violence_path = os.path.join(constants.PATH_VIF_FRAMES,
str(fold + 1),
'NonViolence')
violence_videos = os.listdir(violence_path)
non_violence_videos = os.listdir(non_violence_path)
splitsLen.append(
len(violence_videos) + len(non_violence_videos))
save_file(splitsLen, os.path.join(folder, 'lengths.txt'))
else:
splitsLen = read_file(os.path.join(folder, 'lengths.txt'))
splitsLen = list(map(int, splitsLen))
for i, l in enumerate(splitsLen):
end = np.sum(splitsLen[:(i + 1)])
start = end - splitsLen[i]
test_idx = np.arange(start, end)
train_idx = np.arange(0, start).tolist() + np.arange(
end, len(X)).tolist()
if shuffle:
random.shuffle(train_idx)
random.shuffle(test_idx)
save_file(
train_idx,
os.path.join(folder, 'fold_{}_train.txt'.format(i + 1)))
save_file(
test_idx,
os.path.join(folder, 'fold_{}_test.txt'.format(i + 1)))
for i in range(n_splits):
train_idx = read_file(
os.path.join(folder, 'fold_{}_train.txt'.format(i + 1)))
test_idx = read_file(
os.path.join(folder, 'fold_{}_test.txt'.format(i + 1)))
train_idx = list(map(int, train_idx))
test_idx = list(map(int, test_idx))
yield train_idx, test_idx
elif dataset == 'rwf-2000':
train_idx, test_idx = [], []
for i in range(1):
yield train_idx, test_idx