def im_test(im):
face_info = lib.align(im[:, :, (2, 1, 0)], front_face_detector,
lmark_predictor)
# Samples
if len(face_info) == 0:
logging.warning('No faces are detected.')
prob = -1 # we ignore this case
else:
# Check how many faces in an image
logging.info('{} faces are detected.'.format(len(face_info)))
max_prob = -1
# If one face is fake, the image is fake
for _, point in face_info:
rois = []
for i in range(sample_num):
roi, _ = lib.cut_head([im], point, i)
rois.append(cv2.resize(roi[0], tuple(cfg.IMG_SIZE[:2])))
vis_im(rois, 'tmp/vis.jpg')
prob = solver.test(rois)
prob = np.mean(
np.sort(prob[:, 0])[np.round(sample_num / 2).astype(int):])
if prob >= max_prob:
max_prob = prob
prob = max_prob
return prob
def im_test(net, im, args):
face_info = lib.align(im[:, :, (2, 1, 0)], front_face_detector,
lmark_predictor)
# Samples
if len(face_info) != 1:
prob = -1
else:
_, point = face_info[0]
rois = []
for i in range(sample_num):
roi, _ = lib.cut_head([im], point, i)
rois.append(cv2.resize(roi[0], (args.input_size, args.input_size)))
# vis_ = np.concatenate(rois, 1)
# cv2.imwrite('vis.jpg', vis_)
bgr_mean = np.array([103.939, 116.779, 123.68])
bgr_mean = bgr_mean[np.newaxis, :, np.newaxis, np.newaxis]
bgr_mean = torch.from_numpy(bgr_mean).float().cuda()
rois = torch.from_numpy(np.array(rois)).float().cuda()
rois = rois.permute((0, 3, 1, 2))
prob = net(rois - bgr_mean)
prob = F.softmax(prob, dim=1)
prob = prob.data.cpu().numpy()
prob = 1 - np.mean(
np.sort(prob[:, 0])[np.round(sample_num / 2).astype(int):])
return prob, face_info
def get_batch(self, batch_idx, resize=None):
if batch_idx >= self.batch_num:
raise ValueError("Batch idx must be in range [0, {}].".format(self.batch_num - 1))
imgs = []
names = []
im_path = self.face_img_paths[batch_idx]
im = cv2.imread(str(im_path))
im_name = os.path.basename(im_path).split('.')[0]
_, points = self.face_caches[im_name]
if points is None:
return None
for _ in range(self.sample_num):
# Cut out head region
im_cut, _ = lib.cut_head([im.copy()], points)
im_cut = cv2.resize(im_cut[0], (resize[0], resize[1]))
imgs.append(im_cut)
data = {}
data['images'] = imgs
data['name_list'] = im_name
return data
def get_batch(self, batch_idx, resize=None):
if batch_idx >= self.batch_num:
raise ValueError("Batch idx must be in range [0, {}].".format(self.batch_num - 1))
# Get start and end image index ( counting from 0 )
start_idx = batch_idx * self.batch_size
idx_range = []
for i in range(self.batch_size):
idx_range.append((start_idx + i) % self.data_num)
print('batch index: {}, counting from 0'.format(batch_idx))
imgs = []
labels = []
names = []
for i in idx_range:
im = cv2.imread(str(self.face_img_paths[i]))
im_name = os.path.basename(self.face_img_paths[i]).split('.')[0]
if im_name in self.face_caches:
trans_matrix, point = self.face_caches[im_name]
if point is None:
continue
label = self.annos[im_name]
# label is 1 means this is an authentic image,
if label == 1:
rnd = np.random.uniform()
if rnd < 0.5:
# Affine warp face area back
size = np.arange(64, 128, dtype=np.int32)
c = np.random.randint(0, len(size))
new_im = self._face_blur(im, trans_matrix, size=size[c])
rnd2 = np.random.uniform()
if rnd2 < 0.5:
# Only retain a minimal polygon mask
part_mask = lib.get_face_mask(im.shape[:2], point)
# Select specific blurred part
new_im = self._select_part_to_blur(im, new_im, part_mask)
im = new_im
label = 0
else:
continue
# Cut out head region
ims, _ = lib.cut_head([im], point)
# Augmentation
if self.is_aug:
im = proc_img.aug(ims, random_transform_args=None,
color_rng=[0.8, 1.2])[0]
im = cv2.resize(im, (resize[0], resize[1]))
imgs.append(im)
labels.append(label)
names.append(im_name)
if batch_idx == self.batch_num - 1:
if self.is_shuffle:
idx = np.random.permutation(self.data_num)
self.face_img_paths = [self.face_img_paths[j] for j in idx]
data = {}
data['images'] = imgs
data['images_label'] = labels
data['name_list'] = names
return data