Merge pull request #794 from AnthraX1/master

README.md

@@ -194,7 +194,7 @@ Age model got ± 4.65 MAE; gender model got 97.44% accuracy, 96.29% precision an
 
 **Face Detectors** - [`Demo`](https://youtu.be/GZ2p2hj2H5k)
 
-Face detection and alignment are important early stages of a modern face recognition pipeline. Experiments show that just alignment increases the face recognition accuracy almost 1%. [`OpenCV`](https://sefiks.com/2020/02/23/face-alignment-for-face-recognition-in-python-within-opencv/), [`SSD`](https://sefiks.com/2020/08/25/deep-face-detection-with-opencv-in-python/), [`Dlib`](https://sefiks.com/2020/07/11/face-recognition-with-dlib-in-python/),  [`MTCNN`](https://sefiks.com/2020/09/09/deep-face-detection-with-mtcnn-in-python/), [`RetinaFace`](https://sefiks.com/2021/04/27/deep-face-detection-with-retinaface-in-python/), [`MediaPipe`](https://sefiks.com/2022/01/14/deep-face-detection-with-mediapipe/) and [`YOLOv8 Face`](https://github.com/derronqi/yolov8-face) detectors are wrapped in deepface.
+Face detection and alignment are important early stages of a modern face recognition pipeline. Experiments show that just alignment increases the face recognition accuracy almost 1%. [`OpenCV`](https://sefiks.com/2020/02/23/face-alignment-for-face-recognition-in-python-within-opencv/), [`SSD`](https://sefiks.com/2020/08/25/deep-face-detection-with-opencv-in-python/), [`Dlib`](https://sefiks.com/2020/07/11/face-recognition-with-dlib-in-python/),  [`MTCNN`](https://sefiks.com/2020/09/09/deep-face-detection-with-mtcnn-in-python/), [`RetinaFace`](https://sefiks.com/2021/04/27/deep-face-detection-with-retinaface-in-python/), [`MediaPipe`](https://sefiks.com/2022/01/14/deep-face-detection-with-mediapipe/), [`YOLOv8 Face`](https://github.com/derronqi/yolov8-face) and [`YuNet`](https://github.com/ShiqiYu/libfacedetection) detectors are wrapped in deepface.
 
 <p align="center"><img src="https://raw.githubusercontent.com/serengil/deepface/master/icon/detector-portfolio-v3.jpg" width="95%" height="95%"></p>
 
@@ -209,6 +209,7 @@ backends = [
   'retinaface', 
   'mediapipe',
   'yolov8',
+  'yunet',
 ]
 
 #face verification

deepface/detectors/FaceDetector.py

@@ -10,6 +10,7 @@
     RetinaFaceWrapper,
     MediapipeWrapper,
     YoloWrapper,
+    YunetWrapper,
 )
 
 
@@ -24,6 +25,7 @@ def build_model(detector_backend):
         "retinaface": RetinaFaceWrapper.build_model,
         "mediapipe": MediapipeWrapper.build_model,
         "yolov8": YoloWrapper.build_model,
+        "yunet": YunetWrapper.build_model,
     }
 
     if not "face_detector_obj" in globals():
@@ -67,6 +69,7 @@ def detect_faces(face_detector, detector_backend, img, align=True):
         "retinaface": RetinaFaceWrapper.detect_face,
         "mediapipe": MediapipeWrapper.detect_face,
         "yolov8": YoloWrapper.detect_face,
+        "yunet": YunetWrapper.detect_face,
     }
 
     detect_face_fn = backends.get(detector_backend)

deepface/detectors/YunetWrapper.py

@@ -0,0 +1,77 @@
+import cv2
+import os
+import gdown
+from deepface.detectors import FaceDetector
+from deepface.commons import functions
+
+
+def build_model():
+    url = "https://github.com/opencv/opencv_zoo/raw/main/models/face_detection_yunet/face_detection_yunet_2023mar.onnx"
+    file_name = "face_detection_yunet_2023mar.onnx"
+    home = functions.get_deepface_home()
+    if os.path.isfile(home + f"/.deepface/weights/{file_name}") is False:
+        print(f"{file_name} will be downloaded...")
+        output = home + f"/.deepface/weights/{file_name}"
+        gdown.download(url, output, quiet=False)
+    face_detector = cv2.FaceDetectorYN_create(
+        home + f"/.deepface/weights/{file_name}", "", (0, 0)
+    )
+    return face_detector
+
+
+def detect_face(detector, image, align=True, score_threshold=0.9):
+    # FaceDetector.detect_faces does not support score_threshold parameter.
+    # We can set it via environment variable.
+    score_threshold = os.environ.get("yunet_score_threshold", score_threshold)
+    resp = []
+    detected_face = None
+    img_region = [0, 0, image.shape[1], image.shape[0]]
+    faces = []
+    height, width = image.shape[0], image.shape[1]
+    # resize image if it is too large (Yunet fails to detect faces on large input sometimes)
+    # I picked 640 as a threshold because it is the default value of max_size in Yunet.
+    resized = False
+    if height > 640 or width > 640:
+        r = 640.0 / max(height, width)
+        original_image = image.copy()
+        image = cv2.resize(image, (int(width * r), int(height * r)))
+        height, width = image.shape[0], image.shape[1]
+        resized = True
+    detector.setInputSize((width, height))
+    detector.setScoreThreshold(score_threshold)
+    _, faces = detector.detect(image)
+    if faces is None:
+        return resp
+    for face in faces:
+        """
+        The detection output faces is a two-dimension array of type CV_32F,
+        whose rows are the detected face instances, columns are the location of a face and 5 facial landmarks.
+        The format of each row is as follows:
+        x1, y1, w, h, x_re, y_re, x_le, y_le, x_nt, y_nt, x_rcm, y_rcm, x_lcm, y_lcm,
+        where x1, y1, w, h are the top-left coordinates, width and height of the face bounding box,
+        {x, y}_{re, le, nt, rcm, lcm} stands for the coordinates of right eye, left eye, nose tip, the right corner and left corner of the mouth respectively.
+        """
+        (x, y, w, h, x_re, y_re, x_le, y_le) = list(map(int, face[:8]))
+        if resized:
+            image = original_image
+            x, y, w, h = int(x / r), int(y / r), int(w / r), int(h / r)
+            x_re, y_re, x_le, y_le = (
+                int(x_re / r),
+                int(y_re / r),
+                int(x_le / r),
+                int(y_le / r),
+            )
+        confidence = face[-1]
+        confidence = "{:.2f}".format(confidence)
+        detected_face = image[int(y) : int(y + h), int(x) : int(x + w)]
+        img_region = [x, y, w, h]
+        if align:
+            detected_face = yunet_align_face(detected_face, x_re, y_re, x_le, y_le)
+        resp.append((detected_face, img_region, confidence))
+    return resp
+
+
+# x_re, y_re, x_le, y_le stands for the coordinates of right eye, left eye
+def yunet_align_face(img, x_re, y_re, x_le, y_le):
+    img = FaceDetector.alignment_procedure(img, (x_le, y_le), (x_re, y_re))
+    return img