def faceswap(face_image, face_data, target_image, target_info,
feather_amount=17, color_correct_blur_frac=0.8):
"""
Perform a face swap where face_image and face_landmarks are the faces and
face landmark points for the desired. Target image and target landmarks
point to the face to be changed. Feather amount is the percentage of the
face to feather the face-mask for a nicer transition and
color_correct_blur_frac is the amount of blur to use during color
correction, as a fraction of the pupillary distance.
'aligned_face face_box landmarks imghash'.split()
"""
M = transformation_from_points(
target_info.landmarks[ALIGN_POINTS],
face_data['landmarks'][ALIGN_POINTS]
)
feather_target = feather_amount
feather_face = feather_amount
mask = get_face_mask(
face_image,
face_data['landmarks'],
feather_face
)
warped_mask = warp_im(mask, M, target_image.shape)
combined_mask = np.max([
get_face_mask(target_image, target_info.landmarks, feather_target),
warped_mask
], axis=0)
warped_target_image = warp_im(face_image, M, target_image.shape)
warped_corrected_target_image = correct_colours(
target_image,
warped_target_image,
face_data['landmarks'],
color_correct_blur_frac
)
output_im = target_image * (1.0 - combined_mask) + \
warped_corrected_target_image * combined_mask
return output_im
def _normalize(image, image_size, face_box, landmarks):
H = transformation_from_points(landmarks, ALIGNED_LANDMARKS)
return cv2.warpAffine(image, H[:2], (image_size, image_size), flags=cv2.INTER_AREA)
