def spoofing_face_patch_generate(spoofing_face, save_dir, patch_size=96, display=False):
# nitial
img_transform = ts.Compose([ts.RandomCrop(patch_size)])
img_saved_list = get_file_list(save_dir)
img_patch_num = len(img_saved_list)
img_Image = Image.fromarray(cv2.cvtColor(spoofing_face, cv2.COLOR_BGR2RGB))
data_len = 8
seed_arr = np.arange(data_len)
for i in range(data_len):
img_transform = RandomCrop(size=patch_size, seed=seed_arr[i])
try:
img_patch = img_transform(img_Image)
if display:
img_patch_opencv = cv2.cvtColor(np.array(img_patch), cv2.COLOR_RGB2BGR)
cv2.imshow("patch", img_patch_opencv)
cv2.waitKey(0)
img_patch_num += 1
img_patch_name = "patch" + str(img_patch_num) + ".bmp"
patch_save_path = save_dir + "/" + img_patch_name
img_patch_opencv = cv2.cvtColor(np.array(img_patch), cv2.COLOR_RGB2BGR)
cv2.imwrite(patch_save_path, img_patch_opencv)
except Exception as e:
print(e)
def __init__(self,
living_dir,
spoofing_dir,
balance=True,
data_transform=None):
self.living_path_list = get_file_list(living_dir)
self.spoofing_path_list = get_file_list(spoofing_dir)
# 间隔取样,控制数量
if balance:
self.spoofing_path_list = sorted(self.spoofing_path_list)
balance_factor = int(
np.floor(
len(self.spoofing_path_list) / len(self.living_path_list)))
if balance_factor == 0:
balance_factor = 1
self.spoofing_path_list = self.spoofing_path_list[
0:len(self.spoofing_path_list):balance_factor]
self.img_path_list = self.spoofing_path_list + self.living_path_list
self.data_transform = data_transform
def depth_cnn_test(pre_path, test_dir, label, isface):
'''
:param :
:param pre_path: road to pretrain model
:param test_dir: test img dir
:param isface: img is face img or not, if not need to detect face
:return:
'''
face_detector = FaceDection(model_name='cv2')
model = net_baesd_depth()
state_dict = torch.load(pre_path, map_location='cpu')
model.load_state_dict(state_dict)
model.eval()
time_begin = datetime.datetime.now()
file_list = get_file_list(test_dir)
count = 1
true_num = 1
for file in file_list:
img = cv2.imread(file)
if img is None:
continue
result = depth_cnn_single(model=model,
face_detector=face_detector,
img=img,
isface=isface)
if result is None:
continue
if result == label:
count += 1
true_num += 1
else:
print(file)
count += 1
print(count, true_num, true_num / count)
time_end = datetime.datetime.now()
time_all = time_end - time_begin
print("time_all", time_all.total_seconds())