Update model inference

inference/model_pipeline.py

@@ -9,12 +9,13 @@ from .face_detector import FaceDetector
 
 
 class VSNetModelPipeline:
-    def __init__(self, model, face_detector: FaceDetector, background_resize=720, no_detected_resize=256):
+    def __init__(self, model, face_detector: FaceDetector, background_resize=720, no_detected_resize=256, use_cloning=True):
         self.background_resize = background_resize
         self.no_detected_resize = no_detected_resize
         self.model = model
         self.face_detector = face_detector
         self.mask = self.create_circular_mask(face_detector.target_size, face_detector.target_size)
+        self.use_cloning = use_cloning
 
     @staticmethod
     def create_circular_mask(h, w, power=None, clipping_coef=0.85):
@@ -68,8 +69,12 @@ class VSNetModelPipeline:
             W, H = x2 - x1, y2 - y1
             result_face = cv2.resize(face, (W, H), interpolation=cv2.INTER_LINEAR)
             face_mask = cv2.resize(self.mask, (W, H), interpolation=cv2.INTER_LINEAR)
-            input_face = full_image[y1: y2, x1: x2]
-            full_image[y1: y2, x1: x2] = (result_face * face_mask + input_face * (1 - face_mask)).astype(np.uint8)
+            if self.use_cloning:
+                center = round((x2 + x1) / 2), round((y2 + y1) / 2)
+                full_image = cv2.seamlessClone(result_face, full_image, (face_mask > 0.0).astype(np.uint8) * 255, center, cv2.NORMAL_CLONE)
+            else:
+                input_face = full_image[y1: y2, x1: x2]
+                full_image[y1: y2, x1: x2] = (result_face * face_mask + input_face * (1 - face_mask)).astype(np.uint8)
         return full_image
 
     def __call__(self, img):