Face Detection: replace OpenCV by dlib

REQUIREMENTS.txt

@@ -5,3 +5,4 @@ numpy==1.12.1
 h5py==2.7.0
 statistics
 opencv2-python==3.2.0
+dlib>=19.3.0

src/image_emotion_gender_demo.py

@@ -6,16 +6,17 @@
 
 from utils.datasets import get_labels
 from utils.inference import detect_faces
+from utils.inference import load_detection_model
+from utils.inference import make_face_coordinates
 from utils.inference import draw_text
 from utils.inference import draw_bounding_box
 from utils.inference import apply_offsets
-from utils.inference import load_detection_model
+
 from utils.inference import load_image
 from utils.preprocessor import preprocess_input
 
 # parameters for loading data and images
 image_path = sys.argv[1]
-detection_model_path = '../trained_models/detection_models/haarcascade_frontalface_default.xml'
 emotion_model_path = '../trained_models/emotion_models/fer2013_mini_XCEPTION.102-0.66.hdf5'
 gender_model_path = '../trained_models/gender_models/simple_CNN.81-0.96.hdf5'
 emotion_labels = get_labels('fer2013')
@@ -29,7 +30,7 @@
 emotion_offsets = (0, 0)
 
 # loading models
-face_detection = load_detection_model(detection_model_path)
+face_detection = load_detection_model()
 emotion_classifier = load_model(emotion_model_path, compile=False)
 gender_classifier = load_model(gender_model_path, compile=False)
 
@@ -43,8 +44,12 @@
 gray_image = np.squeeze(gray_image)
 gray_image = gray_image.astype('uint8')
 
-faces = detect_faces(face_detection, gray_image)
-for face_coordinates in faces:
+detected_faces, score, idx = detect_faces(face_detection, gray_image)
+
+for detected_face in detected_faces:
+
+    face_coordinates = make_face_coordinates(detected_face)
+
     x1, x2, y1, y2 = apply_offsets(face_coordinates, gender_offsets)
     rgb_face = rgb_image[y1:y2, x1:x2]
 

src/image_gradcam_demo.py

@@ -10,9 +10,10 @@
 from utils.grad_cam import modify_backprop
 from utils.grad_cam import calculate_guided_gradient_CAM
 from utils.datasets import get_labels
-from utils.inference import detect_faces
 from utils.inference import apply_offsets
+from utils.inference import detect_faces
 from utils.inference import load_detection_model
+from utils.inference import make_face_coordinates
 from utils.preprocessor import preprocess_input
 from utils.inference import draw_bounding_box
 from utils.inference import load_image
@@ -35,20 +36,21 @@
 color = (0, 255, 0)
 
 # loading models
-detection_model_path = '../trained_models/detection_models/haarcascade_frontalface_default.xml'
 model = load_model(model_filename, compile=False)
 target_size = model.input_shape[1:3]
-face_detection = load_detection_model(detection_model_path)
+face_detection = load_detection_model()
 
 # loading images
 rgb_image = load_image(image_path, grayscale=False)
 gray_image = load_image(image_path, grayscale=True)
 gray_image = np.squeeze(gray_image)
 gray_image = gray_image.astype('uint8')
-faces = detect_faces(face_detection, gray_image)
+detected_faces, score, idx = detect_faces(face_detection, gray_image)
 
 # start prediction for every image
-for face_coordinates in faces:
+for detected_face in detected_faces:
+
+    face_coordinates = make_face_coordinates(detected_face)
 
     x1, x2, y1, y2 = apply_offsets(face_coordinates, offsets)
     rgb_face = rgb_image[y1:y2, x1:x2]

src/utils/inference.py

@@ -1,4 +1,5 @@
 import cv2
+import dlib
 import matplotlib.pyplot as plt
 import numpy as np
 from keras.preprocessing import image
@@ -7,12 +8,18 @@ def load_image(image_path, grayscale=False, target_size=None):
     pil_image = image.load_img(image_path, grayscale, target_size)
     return image.img_to_array(pil_image)
 
-def load_detection_model(model_path):
-    detection_model = cv2.CascadeClassifier(model_path)
-    return detection_model
+def load_detection_model():
+    return dlib.get_frontal_face_detector()
 
 def detect_faces(detection_model, gray_image_array):
-    return detection_model.detectMultiScale(gray_image_array, 1.3, 5)
+    return detection_model.run(gray_image_array, 0, 0)
+
+def make_face_coordinates(detected_face):
+    x = detected_face.left()
+    y = detected_face.top()
+    width = detected_face.right() - detected_face.left()
+    height = detected_face.bottom() - detected_face.top()
+    return [x, y, width, height]
 
 def draw_bounding_box(face_coordinates, image_array, color):
     x, y, w, h = face_coordinates

src/video_emotion_color_demo.py

@@ -10,10 +10,10 @@
 from utils.inference import draw_bounding_box
 from utils.inference import apply_offsets
 from utils.inference import load_detection_model
+from utils.inference import make_face_coordinates
 from utils.preprocessor import preprocess_input
 
 # parameters for loading data and images
-detection_model_path = '../trained_models/detection_models/haarcascade_frontalface_default.xml'
 emotion_model_path = '../trained_models/emotion_models/fer2013_mini_XCEPTION.102-0.66.hdf5'
 emotion_labels = get_labels('fer2013')
 
@@ -22,7 +22,7 @@
 emotion_offsets = (20, 40)
 
 # loading models
-face_detection = load_detection_model(detection_model_path)
+face_detection = load_detection_model()
 emotion_classifier = load_model(emotion_model_path, compile=False)
 
 # getting input model shapes for inference
@@ -38,10 +38,11 @@
     bgr_image = video_capture.read()[1]
     gray_image = cv2.cvtColor(bgr_image, cv2.COLOR_BGR2GRAY)
     rgb_image = cv2.cvtColor(bgr_image, cv2.COLOR_BGR2RGB)
-    faces = detect_faces(face_detection, gray_image)
+    detected_faces, score, idx = detect_faces(face_detection, gray_image)
 
-    for face_coordinates in faces:
+    for detected_face in detected_faces:
 
+        face_coordinates = make_face_coordinates(detected_face)
         x1, x2, y1, y2 = apply_offsets(face_coordinates, emotion_offsets)
         gray_face = gray_image[y1:y2, x1:x2]
         try:

src/video_emotion_gender_demo.py

@@ -10,10 +10,10 @@
 from utils.inference import draw_bounding_box
 from utils.inference import apply_offsets
 from utils.inference import load_detection_model
+from utils.inference import make_face_coordinates
 from utils.preprocessor import preprocess_input
 
 # parameters for loading data and images
-detection_model_path = '../trained_models/detection_models/haarcascade_frontalface_default.xml'
 emotion_model_path = '../trained_models/emotion_models/fer2013_mini_XCEPTION.102-0.66.hdf5'
 gender_model_path = '../trained_models/gender_models/simple_CNN.81-0.96.hdf5'
 emotion_labels = get_labels('fer2013')
@@ -26,7 +26,7 @@
 emotion_offsets = (20, 40)
 
 # loading models
-face_detection = load_detection_model(detection_model_path)
+face_detection = load_detection_model()
 emotion_classifier = load_model(emotion_model_path, compile=False)
 gender_classifier = load_model(gender_model_path, compile=False)
 
@@ -46,9 +46,11 @@
     bgr_image = video_capture.read()[1]
     gray_image = cv2.cvtColor(bgr_image, cv2.COLOR_BGR2GRAY)
     rgb_image = cv2.cvtColor(bgr_image, cv2.COLOR_BGR2RGB)
-    faces = detect_faces(face_detection, gray_image)
+    detected_faces, score, idx = detect_faces(face_detection, gray_image)
 
-    for face_coordinates in faces:
+    for detected_face in detected_faces:
+
+        face_coordinates = make_face_coordinates(detected_face)
 
         x1, x2, y1, y2 = apply_offsets(face_coordinates, gender_offsets)
         rgb_face = rgb_image[y1:y2, x1:x2]

src/video_gradcam_demo.py

@@ -11,6 +11,7 @@
 from utils.inference import detect_faces
 from utils.inference import apply_offsets
 from utils.inference import load_detection_model
+from utils.inference import make_face_coordinates
 from utils.preprocessor import preprocess_input
 from utils.inference import draw_bounding_box
 from utils.datasets import get_class_to_arg
@@ -39,9 +40,8 @@
 guided_model = modify_backprop(model, 'GuidedBackProp', task)
 saliency_function = compile_saliency_function(guided_model, 'conv2d_7')
 
-# parameters for loading data and images 
-detection_model_path = '../trained_models/detection_models/haarcascade_frontalface_default.xml'
-face_detection = load_detection_model(detection_model_path)
+# parameters for loading data and images
+face_detection = load_detection_model()
 color = (0, 255, 0)
 
 # getting input model shapes for inference
@@ -57,10 +57,11 @@
     bgr_image = video_capture.read()[1]
     gray_image = cv2.cvtColor(bgr_image, cv2.COLOR_BGR2GRAY)
     rgb_image = cv2.cvtColor(bgr_image, cv2.COLOR_BGR2RGB)
-    faces = detect_faces(face_detection, gray_image)
+    detected_faces, score, idx = detect_faces(face_detection, gray_image)
 
-    for face_coordinates in faces:
+    for detected_face in detected_faces:
 
+        face_coordinates = make_face_coordinates(detected_face)
         x1, x2, y1, y2 = apply_offsets(face_coordinates, offsets)
         gray_face = gray_image[y1:y2, x1:x2]
         try: