[feat] : Migration from Mediapipe Solutions to Mediapipe Tasks Framework

eyeGestures/eye.py

@@ -13,8 +13,12 @@
 class Eye:
     """Class storing data related and representing a eye"""
 
-    LEFT_EYE_KEYPOINTS = np.array(list(mp.solutions.face_mesh.FACEMESH_LEFT_EYE))[:, 0]
-    RIGHT_EYE_KEYPOINTS = np.array(list(mp.solutions.face_mesh.FACEMESH_RIGHT_EYE))[:, 0]
+    LEFT_EYE_KEYPOINTS = np.array([
+        33, 7, 163, 144, 145, 153, 154, 155, 133, 246, 161, 160, 159, 158, 157, 173
+    ])
+    RIGHT_EYE_KEYPOINTS = np.array([
+        263, 249, 390, 373, 374, 380, 381, 382, 362, 398, 384, 385, 386, 387, 388, 466
+    ])
     LEFT_EYE_PUPIL_KEYPOINT = [473]
     RIGHT_EYE_PUPIL_KEYPOINT = [468]
 

eyeGestures/face.py

@@ -4,6 +4,11 @@
 
 import cv2
 import mediapipe as mp
+
+from mediapipe.tasks import python
+from mediapipe.tasks.python import vision
+from mediapipe import Image, ImageFormat
+
 import numpy as np
 import numpy.typing as npt
 
@@ -14,21 +19,31 @@ class FaceFinder:
     """Class helping finding face"""
 
     def __init__(self) -> None:
-        self.mp_face_mesh = mp.solutions.face_mesh.FaceMesh(
-            refine_landmarks=True,
-            static_image_mode=False,
-            min_detection_confidence=0.5,
-            min_tracking_confidence=0.5,
+        BaseOptions = python.BaseOptions
+        FaceLandmarker = vision.FaceLandmarker
+        FaceLandmarkerOptions = vision.FaceLandmarkerOptions
+        VisionRunningMode = vision.RunningMode
+        options = FaceLandmarkerOptions(
+            base_options = BaseOptions(
+                model_asset_path = r"C:\Users\vvams\PycharmProjects\EyeGestures\eyeGestures\task_model\face_landmarker.task"
+            ),
+            running_mode = VisionRunningMode.IMAGE,
+            num_faces = 1,
+            min_face_detection_confidence=0.5,
+            min_tracking_confidence=0.5
         )
+        self.mp_face_mesh = FaceLandmarker.create_from_options(options)
 
     def find(self, image: cv2.typing.MatLike) -> Optional[Any]:
         """Find face mesh"""
 
         assert len(image.shape) > 2
 
         try:
-            face_mesh = self.mp_face_mesh.process(cv2.cvtColor(image, cv2.COLOR_BGR2RGB))
-            if face_mesh.multi_face_landmarks is None:
+            image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
+            mp_image = Image(image_format = ImageFormat.SRGB, data = image_rgb)
+            face_mesh = self.mp_face_mesh.detect(mp_image)
+            if face_mesh.face_landmarks is None:
                 return None
 
             return face_mesh
@@ -78,11 +93,10 @@ def get_landmarks(self) -> Optional[npt.NDArray[np.float64]]:
 
     def _landmarks(self, face: Any) -> npt.NDArray[np.float64]:
 
-        __complex_landmark_points = face.multi_face_landmarks
-        __complex_landmarks = __complex_landmark_points[0].landmark
+        __complex_landmark_points = face.face_landmarks[0]
 
         __face_landmarks = []
-        for landmark in __complex_landmarks:
+        for landmark in __complex_landmark_points:
             __face_landmarks.append((landmark.x * self.image_w, landmark.y * self.image_h))
 
         return np.array(__face_landmarks)

eyeGestures/gazeEstimator.py

@@ -112,7 +112,7 @@ def estimate(
         if not face_mesh:
             return None
 
-        if face_mesh.multi_face_landmarks:
+        if face_mesh.face_landmarks:
             self.face.process(image, face_mesh)
 
         context = self.GContext.get(

eyeGestures/processing.py

@@ -32,8 +32,8 @@ def append(self, pupil: Tuple[int, int], landmarks: npt.NDArray[np.float64], pup
         self.min_y = np.min(self.landmarks[:, 1]) - margin
         self.max_y = np.max(self.landmarks[:, 1]) + margin
 
-        assert self.pupil[0] > self.min_x
-        assert self.pupil[1] > self.min_y
+        # assert self.pupil[0] > self.min_x
+        # assert self.pupil[1] > self.min_y
 
         width = self.max_x - self.min_x
         height = (self.max_y - self.min_y) / 2