def swap(sF, iF, sT, iT):
warp_2d = True ##!!
correct_color = True ##!!
# Select src face
src_points, src_shape, src_face = select_face(sF, iF)
viewOriginal(src_face)
# Select dst face
dst_points, dst_shape, dst_face = select_face(sT, iT)
viewOriginal(dst_face)
w, h = dst_face.shape[:2]
### Warp Image
if not warp_2d:
## 3d warp
warped_src_face = warp_image_3d(src_face, src_points[:48],
dst_points[:48], (w, h))
else:
## 2d warp
src_mask = mask_from_points(src_face.shape[:2], src_points)
src_face = apply_mask(src_face, src_mask)
# Correct Color for 2d warp
if correct_color:
warped_dst_img = warp_image_3d(dst_face, dst_points[:48],
src_points[:48], src_face.shape[:2])
src_face = correct_colours(warped_dst_img, src_face, src_points)
# Warp
warped_src_face = warp_image_2d(
src_face, transformation_from_points(dst_points, src_points),
(w, h, 3))
## Mask for blending
mask = mask_from_points((w, h), dst_points)
mask_src = np.mean(warped_src_face, axis=2) > 0
mask = np.asarray(mask * mask_src, dtype=np.uint8)
## Correct color
if not warp_2d and correct_color:
warped_src_face = apply_mask(warped_src_face, mask)
dst_face_masked = apply_mask(dst_face, mask)
warped_src_face = correct_colours(dst_face_masked, warped_src_face,
dst_points)
## Shrink the mask
kernel = np.ones((10, 10), np.uint8)
mask = cv2.erode(mask, kernel, iterations=1)
##Poisson Blending
r = cv2.boundingRect(mask)
center = ((r[0] + int(r[2] / 2), r[1] + int(r[3] / 2)))
output = cv2.seamlessClone(warped_src_face, dst_face, mask, center,
cv2.NORMAL_CLONE)
x, y, w, h = dst_shape
dst_img_cp = iT.copy()
dst_img_cp[y:y + h, x:x + w] = output
output = dst_img_cp
return output
def apply_facial_features(src, dst, out, warp_2d = False, correct_color = True):
# Read images
src_img = cv2.imread(src)
dst_img = cv2.imread(dst)
# Select src face
src_points, src_shape, src_face = select_face(src_img)
# Select dst face
dst_points, dst_shape, dst_face = select_face(dst_img)
w, h = dst_face.shape[:2]
# Warp Image
if not warp_2d:
# 3d warp
warped_src_face = warp_image_3d(src_face, src_points[:48], dst_points[:48], (w, h))
else:
# 2d warp
src_mask = mask_from_points(src_face.shape[:2], src_points)
src_face = apply_mask(src_face, src_mask)
# Correct Color for 2d warp
if correct_color:
warped_dst_img = warp_image_3d(dst_face, dst_points[:48], src_points[:48], src_face.shape[:2])
src_face = correct_colours(warped_dst_img, src_face, src_points)
# Warp
warped_src_face = warp_image_2d(src_face, transformation_from_points(dst_points, src_points), (w, h, 3))
# Mask for blending
mask = mask_from_points((w, h), dst_points)
mask_src = np.mean(warped_src_face, axis=2) > 0
mask = np.asarray(mask * mask_src, dtype=np.uint8)
# Correct color
if not warp_2d and correct_color:
warped_src_face = apply_mask(warped_src_face, mask)
dst_face_masked = apply_mask(dst_face, mask)
warped_src_face = correct_colours(dst_face_masked, warped_src_face, dst_points)
# Shrink the mask
kernel = np.ones((10, 10), np.uint8)
mask = cv2.erode(mask, kernel, iterations=1)
# Poisson Blending
r = cv2.boundingRect(mask)
center = ((r[0] + int(r[2] / 2), r[1] + int(r[3] / 2)))
output = cv2.seamlessClone(warped_src_face, dst_face, mask, center, cv2.NORMAL_CLONE)
x, y, w, h = dst_shape
dst_img_cp = dst_img.copy()
dst_img_cp[y:y + h, x:x + w] = output
output = dst_img_cp
dir_path = os.path.dirname(out)
if not os.path.isdir(dir_path):
os.makedirs(dir_path)
cv2.imwrite(out, output)
h, w = dst_face.shape[:2]
### Warp Image
if not args.warp_2d:
## 3d warp
warped_src_face = warp_image_3d(src_face, src_points[:48], dst_points[:48], (h, w))
else:
## 2d warp
src_mask = mask_from_points(src_face.shape[:2], src_points)
src_face = apply_mask(src_face, src_mask)
# Correct Color for 2d warp
if args.correct_color:
warped_dst_img = warp_image_3d(dst_face, dst_points[:48], src_points[:48], src_face.shape[:2])
src_face = correct_colours(warped_dst_img, src_face, src_points)
# Warp
warped_src_face = warp_image_2d(src_face, transformation_from_points(dst_points, src_points), (h, w, 3))
## Mask for blending
mask = mask_from_points((h, w), dst_points)
mask_src = np.mean(warped_src_face, axis=2) > 0
mask = np.asarray(mask*mask_src, dtype=np.uint8)
## Correct color
if not args.warp_2d and args.correct_color:
warped_src_face = apply_mask(warped_src_face, mask)
dst_face_masked = apply_mask(dst_face, mask)
warped_src_face = correct_colours(dst_face_masked, warped_src_face, dst_points)
## Shrink the mask
kernel = np.ones((10, 10), np.uint8)
mask = cv2.erode(mask, kernel, iterations=1)
def insert_face(result, CROWD, scale):
if result is None:
return None, None
result_bboxs = []
for faces in result:
dst_points, dst_shape, dst_face = faces[0]
src_points, src_shape, src_face = faces[1]
w, h = dst_face.shape[:2]
# 3d warp if face is big enough
WARP_2D = True
if (h / CROWD.shape[0]) >= FACE_FRAC_3D:
WARP_2D = False
### Warp Image
if not WARP_2D:
## 3d warp
warped_src_face = warp_image_3d(src_face, src_points[:MAX_POINTS],
dst_points[:MAX_POINTS], (w, h))
else:
## 2d warp
src_mask = mask_from_points(src_face.shape[:2],
src_points,
radius=2)
src_face = apply_mask(src_face, src_mask)
# Correct Color for 2d warp
if CORRECT_COLOR:
warped_dst_img = warp_image_3d(dst_face,
dst_points[:MAX_POINTS],
src_points[:MAX_POINTS],
src_face.shape[:2])
src_face = correct_colours(warped_dst_img, src_face,
src_points)
# Warp
warped_src_face = warp_image_2d(
src_face, transformation_from_points(dst_points, src_points),
(w, h, 3))
## Mask for blending
mask = mask_from_points((w, h), dst_points, radius=2)
mask_src = np.mean(warped_src_face, axis=2) > 0
mask = np.asarray(mask * mask_src, dtype=np.uint8)
## Correct color
if not WARP_2D and CORRECT_COLOR:
warped_src_face = apply_mask(warped_src_face, mask)
dst_face_masked = apply_mask(dst_face, mask)
warped_src_face = correct_colours(dst_face_masked, warped_src_face,
dst_points)
##Poisson Blending
r = cv2.boundingRect(mask)
center = ((r[0] + int(r[2] / 2), r[1] + int(r[3] / 2)))
output = cv2.seamlessClone(warped_src_face, dst_face, mask, center,
cv2.NORMAL_CLONE)
x, y, w, h = dst_shape
result_bboxs.append([x, y, x + w, y + h])
CROWD[y:y + h, x:x + w] = output
return CROWD, result_bboxs
