Add YOLOv8n-face, with confidence

deepface/detectors/FaceDetector.py

@@ -9,11 +9,11 @@ from deepface.detectors import (
     MtcnnWrapper,
     RetinaFaceWrapper,
     MediapipeWrapper,
+    Yolov8nfaceWrapper,
 )
 
 
 def build_model(detector_backend):
-
     global face_detector_obj  # singleton design pattern
 
     backends = {
@@ -23,6 +23,7 @@ def build_model(detector_backend):
         "mtcnn": MtcnnWrapper.build_model,
         "retinaface": RetinaFaceWrapper.build_model,
         "mediapipe": MediapipeWrapper.build_model,
+        "yolov8n-face": Yolov8nfaceWrapper.build_model,
     }
 
     if not "face_detector_obj" in globals():
@@ -42,7 +43,6 @@ def build_model(detector_backend):
 
 
 def detect_face(face_detector, detector_backend, img, align=True):
-
     obj = detect_faces(face_detector, detector_backend, img, align)
 
     if len(obj) > 0:
@@ -50,12 +50,12 @@ def detect_face(face_detector, detector_backend, img, align=True):
     else:  # len(obj) == 0
         face = None
         region = [0, 0, img.shape[1], img.shape[0]]
+        confidence = 0
 
     return face, region, confidence
 
 
 def detect_faces(face_detector, detector_backend, img, align=True):
-
     backends = {
         "opencv": OpenCvWrapper.detect_face,
         "ssd": SsdWrapper.detect_face,
@@ -63,6 +63,7 @@ def detect_faces(face_detector, detector_backend, img, align=True):
         "mtcnn": MtcnnWrapper.detect_face,
         "retinaface": RetinaFaceWrapper.detect_face,
         "mediapipe": MediapipeWrapper.detect_face,
+        "yolov8n-face": Yolov8nfaceWrapper.detect_face,
     }
 
     detect_face_fn = backends.get(detector_backend)
@@ -76,7 +77,6 @@ def detect_faces(face_detector, detector_backend, img, align=True):
 
 
 def alignment_procedure(img, left_eye, right_eye):
-
     # this function aligns given face in img based on left and right eye coordinates
 
     left_eye_x, left_eye_y = left_eye
@@ -104,7 +104,6 @@ def alignment_procedure(img, left_eye, right_eye):
     # apply cosine rule
 
     if b != 0 and c != 0:  # this multiplication causes division by zero in cos_a calculation
-
         cos_a = (b * b + c * c - a * a) / (2 * b * c)
         angle = np.arccos(cos_a)  # angle in radian
         angle = (angle * 180) / math.pi  # radian to degree

deepface/detectors/OpenCvWrapper.py

@@ -45,6 +45,7 @@ def detect_face(detector, img, align=True):
     img_region = [0, 0, img.shape[1], img.shape[0]]
 
     faces = []
+    scores = []
     try:
         # faces = detector["face_detector"].detectMultiScale(img, 1.3, 5)
 
@@ -52,19 +53,21 @@ def detect_face(detector, img, align=True):
         faces, _, scores = detector["face_detector"].detectMultiScale3(
             img, 1.1, 10, outputRejectLevels=True
         )
-    except:
+    except Exception:  # pylint: disable=broad-except
-        pass
+        # except alone is too broad and will catch keyboard interrupts
+        import traceback
 
-    if len(faces) > 0:
+        print(traceback.format_exc())
-        for (x, y, w, h), confidence in zip(faces, scores):
-            detected_face = img[int(y) : int(y + h), int(x) : int(x + w)]
 
-            if align:
+    for (x, y, w, h), confidence in zip(faces, scores):
-                detected_face = align_face(detector["eye_detector"], detected_face)
+        detected_face = img[int(y) : int(y + h), int(x) : int(x + w)]
 
-            img_region = [x, y, w, h]
+        if align:
+            detected_face = align_face(detector["eye_detector"], detected_face)
 
-            resp.append((detected_face, img_region, confidence))
+        img_region = [x, y, w, h]
+
+        resp.append((detected_face, img_region, confidence))
 
     return resp
 

deepface/detectors/Yolov8nfaceWrapper.py

@@ -0,0 +1,42 @@
+def build_model():
+    import gdown
+    import os
+
+    from ultralytics import YOLO
+
+    from deepface.commons.functions import get_deepface_home
+
+    weights_path = f"{get_deepface_home()}/.deepface/weights/yolov8n-face.pt"
+
+    if not os.path.isfile(weights_path):
+        url = "https://drive.google.com/uc?id=1qcr9DbgsX3ryrz2uU8w4Xm3cOrRywXqb"
+        gdown.download(url, weights_path, quiet=False)
+        print("Downloaded YOLO model yolo8vn-face.pt")
+
+    # return face_detector
+    return YOLO(weights_path)
+
+
+def detect_face(face_detector, img, align=False):
+    resp = []
+    confidence = -1
+    detected_face = None
+
+    # if align:
+    #     raise NotImplementedError("`align` is not implemented for Yolov8Wrapper")
+
+    results = face_detector.predict(img, verbose=False, show=True, conf=0.25)[0]
+
+    for result in results:
+        x, y, w, h = result.boxes.xywh.tolist()[0]
+        confidence = result.boxes.conf.tolist()[0]
+        # print(f"Confidence: {confidence}, x: {x}, y: {y}, w: {w}, h: {h}")
+
+        # print landmarks
+        print(result.keypoints.tolist())
+        # change to top left corner, width, height
+        x, y, w, h = int(x - w / 2), int(y - h / 2), int(w), int(h)
+        detected_face = img[y : y + h, x : x + w].copy()
+        resp.append((detected_face, [x, y, w, h], confidence))
+
+    return resp