fix weight relative path in model

model/inceptionv3.py

@@ -14,7 +14,8 @@ def __init__(self):
         base = InceptionV3(
             input_shape=(140, 140, 3),
             include_top=False,
-            weights='weight/inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5')
+            weights=os.path.dirname(
+                __file__)+'/weight/inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5')
         top_layer = GlobalAveragePooling2D(name='avg_pool')(base.output)
         gender_layer = Dense(2, activation='softmax', name='gender_prediction')(top_layer)
         age_layer = Dense(101, activation='softmax', name='age_prediction')(top_layer)

model/insightface.py

@@ -93,7 +93,8 @@ class InsightFace:
     def __init__(self):
         ctx = mx.gpu(0)
         self.image_size = (112, 112)
-        self.ga_model = get_model(ctx, self.image_size, 'weight/model-r34-age/model,0', 'fc1')
+        self.ga_model = get_model(ctx, self.image_size, os.path.dirname(
+            __file__)+'/weight/model-r34-age/model,0', 'fc1')
         self.threshold = 1.24
         self.det_minsize = 50
         self.det_threshold = [0.6, 0.7, 0.8]

model/mobilenetv2.py

@@ -1,4 +1,5 @@
 import numpy as np
+import os
 from keras.models import Model
 from keras.layers import Dense, GlobalAveragePooling2D
 from keras.applications.mobilenetv2 import MobileNetV2
@@ -11,10 +12,8 @@ class AgenderNetMobileNetV2(Model):
 
     def __init__(self):
         self.input_size = 96
-        base = MobileNetV2(
-            input_shape=(96, 96, 3),
-            include_top=False,
-            weights='weight/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.0_96_no_top.h5')
+        base = MobileNetV2(input_shape=(96, 96, 3), include_top=False, weights=os.path.dirname(
+            __file__)+'/weight/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.0_96_no_top.h5')
         top_layer = GlobalAveragePooling2D()(base.output)
         gender_layer = Dense(2, activation='softmax', name='gender_prediction')(top_layer)
         age_layer = Dense(101, activation='softmax', name='age_prediction')(top_layer)

utils/image.py

@@ -0,0 +1,70 @@
+import os
+import cv2
+import dlib
+import numpy as np
+
+
+def resize_square_image(image: np.ndarray, size: int =140):
+    """
+    Resize image and make it square
+
+    Parameters
+    ----------
+    image   : numpy array -> with dtype uint8 and shape (W, H, 3)
+        Image to be resized
+    size        : int
+        Size of image to be returned
+
+    Returns
+    ----------
+    resized    : numpy array -> with dtype uint8 and shape (size, size, 3)
+        Resized image
+    """
+    BLACK = [0, 0, 0]
+    h = image.shape[0]
+    w = image.shape[1]
+    if w < h:
+        border = h-w
+        image = cv2.copyMakeBorder(
+            image, 0, 0, border, 0, cv2.BORDER_CONSTANT, value=BLACK)
+    else:
+        border = w-h
+        image = cv2.copyMakeBorder(
+            image, border, 0, 0, 0, cv2.BORDER_CONSTANT, value=BLACK)
+    resized = cv2.resize(image, (size, size), interpolation=cv2.INTER_CUBIC)
+    return resized
+
+
+def align_faces(image: np.ndarray, padding: float=0.4, size: int=140, predictor_path: str=os.path.dirname(
+        os.path.dirname(__file__)) + '/model/shape_predictor_5_face_landmarks.dat'):
+    """
+    Get aligned faces from image if face is detected, else just resize
+    Parameters
+    ----------
+    image   : numpy array -> with dtype uint8 and shape (W, H, 3)
+        Image to be processed
+    padding : float
+        Padding for aligned face
+    size    : int
+        Size of image to be returned
+    path    : string
+        Path to dlib facial landmark detector
+    Returns
+    ----------
+    result    : numpy array -> with dtype uint8 and shape (size, size, 3)
+        Processed image
+    """
+    detector = dlib.get_frontal_face_detector()
+    predictor = dlib.shape_predictor(predictor_path)
+    rects = detector(image, 1)
+    result = None
+
+    # if detect exactly 1 face, get aligned face
+    if len(rects) == 1:
+        shape = predictor(image, rects[0])
+        result = dlib.get_face_chip(image, shape, padding=padding, size=size)
+
+    # else use resized full image
+    else:
+        result = resize_square_image(image, size)
+    return result

utils/stream.py

@@ -29,7 +29,7 @@ def update(self):
     def elapsed(self):
         # return the total number of seconds between the start and
         # end interval
-        return (self._end - self._start).total_seconds()
+        return (datetime.datetime.now() - self._start).total_seconds()
 
     def fps(self):
         # compute the (approximate) frames per second