resize issues resolved; sources included; multiple style images

StyleTransfer.py

@@ -64,8 +64,14 @@ def main():
     parser = build_parser()
     args = parser.parse_args()
     content_image = load_img_preprocess(args.content)
-    style_image = load_img_preprocess(args.style)
-
+    #style_image = load_img_preprocess(args.style)
+	style_paths = args.style
+	style_images = []
+
+#     make an array and add all style images produced from style paths given as input
+    for i in style_paths:
+#         call the load and preprocess function to load the style images from the given paths
+        style_images.append(load_img_preprocess(i))
     """
     get the model from keras basically lets us extract the layers
     and their corresponding intermediate and batch outputs
@@ -96,12 +102,18 @@ def main():
     for layer in main_model.layers:
         layer.trainable = False
 
-    style_features = [style_layer[0]
-                      for style_layer in main_model(style_image)[:5]]
+	style_features = []
+
+	for i in style_images:
+        style_features.append([style_layer[0]
+                       for style_layer in main_model(i)[:5]])
+
     content_features = [content_layer[0]
                         for content_layer in main_model(content_image)[5:]]
-
-    gram_style_features = [gram_matrix(style_feature) for style_feature in style_features]
+	gram_style_features = []
+
+    for i in style_features:
+        gram_style_features.append([gram_matrix(style_feature) for style_feature in i])
 
     # Set initial image
     init_image = load_img_preprocess(args.content)
@@ -142,8 +154,9 @@ def main():
 
 
             style_layer_norm = 1/float(5)
-            for out, inter in zip(gram_style_features, style_out):
-                stylePoints = stylePoints + style_layer_norm*get_style_loss(inter[0],out)
+            for grams in gram_style_features:
+                for out, inter in zip(grams, style_out):
+                    stylePoints = stylePoints + style_layer_norm*get_style_loss(inter[0],out)
 
             contentPoints = contentPoints*contentPoints
             stylePoints = stylePoints*stylePoints
@@ -170,22 +183,52 @@ def main():
     plt.imsave(output_name, display_img)
     plt.show()
 
-
+# referece
+# citing URL: https://keras.io/applications/#vgg19
 def load_img_preprocess(image_path):
     img_str = tf.read_file(image_path)
+
     img_decode = tf.image.decode_jpeg(img_str, 3)
+
     img = tf.cast(img_decode, tf.float32)
+
     dim =512.0
+
     height = tf.to_float(tf.shape(img)[1])
+
     width = tf.to_float(tf.shape(img)[0])
-    scale = tf.cond(tf.greater(height, width), lambda: dim/width, lambda: dim/height)
+
+    print('this is the old height and width ', height, width)
+    scale = tf.cond(tf.greater(height, width), lambda: dim/width , lambda: dim/height)
+    print('this is the scale ', scale)
+
     newHeight = tf.to_int32(height * scale)
     newWidth = tf.to_int32(width * scale)
+    print('newheight and new width', newHeight, newWidth)
+
     img = tf.image.resize_images(img, [newHeight, newWidth])
+
+    """VGG_MEAN = [123.68, 116.78, 103.94]  # This is R-G-B for Imagenet
+
+    img = tf.random_crop(img, [224, 224, 3])
+    means = tf.reshape(tf.constant(VGG_MEAN), [1, 1, 3])
+    img = img - means
+    """
     img = np.expand_dims(img, axis=0)
-    VGG_MEAN = [123.68, 116.78, 103.94] #RGB Values for ImageNet
+
+    VGG_MEAN = [123.68, 116.78, 103.94]
+
     means = tf.reshape(tf.constant(VGG_MEAN), [1, 1, 3])
     img = img - means
+    max_dim = 512
+    img = Image.open(image_path)
+    norm = max(img.size)
+    scale = max_dim/norm
+    img = img.resize((round(img.size[0]*scale), round(img.size[1]*scale)), Image.ANTIALIAS)
+    img = kp_image.img_to_array(img)
+    img = np.expand_dims(img, axis=0)
+    img = tf.keras.applications.vgg19.preprocess_input(img)
+
     return img