Tunend intital values and also the range

app.py

@@ -159,8 +159,8 @@ def fit_style_transfer(
 
 
 def main():
-    content_image = gr.Image(type="filepath", label="Content Image", shape=(512, 512))
-    style_image = gr.Image(type="filepath", label="Style Image", shape=(512, 512))
+    content_image = gr.Image(type="filepath", label="Content Image")
+    style_image = gr.Image(type="filepath", label="Style Image")
 
     extractor = gr.Dropdown(
         ["inception_v3", "vgg19", "resnet50", "mobilenet_v2"],
@@ -180,7 +180,7 @@ def main():
     n_style_layers = gr.Slider(
         1,
         5,
-        value=2,
+        value=3,
         step=1,
         label="Style Layers",
     )
@@ -191,34 +191,31 @@ def main():
 
     steps_per_epoch = gr.Slider(
         1,
-        10,
+        20,
         value=5,
         step=1,
         label="Steps Per Epoch",
     )
 
     style_weight = gr.Slider(
-        1e-4,
-        1e-1,
-        value=5e-2,
-        step=1e-4,
+        0,
+        1,
+        value=0.1,
         label="Style Weight",
     )
 
     content_weight = gr.Slider(
-        1e-3,
-        1e-1,
-        value=3e-2,
-        step=1e-4,
+        0,
+        1,
+        value=1,
         label="Content Weight",
     )
 
     var_weight = gr.Slider(
         0,
-        3,
-        value=1.5,
-        step=0.2,
-        label="Total Variation Weight",
+        0.8,
+        value=0,
+        label="Variation Weight",
     )
 
     inputs = [
@@ -237,49 +234,34 @@ def main():
 
     examples = [
         [
-            "examples/style_1.jpg",
-            "examples/content_1.jpg",
+            "examples/van_gogh.jpg",
+            "examples/scene.jpg",
             "inception_v3",
-            3,
-            1,
             5,
-            80,
+            5,
             10,
-            2e-2,
-            1e-2,
-            1.5,
-        ],
-        [
-            "examples/style_2.jpg",
-            "examples/content_2.jpg",
-            "inception_v3",
-            3,
-            2,
-            4,
-            60,
+            40,
             10,
-            5e-2,
-            3e-2,
-            1.5,
+            0.1,
+            1,
+            0.01,
         ],
-        
         [
-            "examples/style_3.jpg",
-            "examples/content_3.jpg",
-            "inception_v3",
-            3,
-            1,
-            4,
-            60,
+            "examples/painting.jpg",
+            "examples/swan.jpg",
+            "vgg19",
+            5,
+            5,
             10,
-            5e-2,
-            3e-2,
-            1.5,
+            40,
+            10,
+            0.1,
+            1,
+            0.01,
         ]
-
     ]
 
-    output_image = gr.Image(type="numpy", label="Output Image", shape=(512, 512))
+    output_image = gr.Image(type="numpy", label="Output Image")
 
     style_loss = gr.Number(label="Current Style Loss")
 
@@ -291,6 +273,10 @@ def main():
 
     curr_step = gr.Number(label="Current Step")
 
+    title = "Neural Style Transfer"
+
+    description = """### This app uses a neural network to transfer the style of one image to another. \n### The `style image` is the image whose style you want to transfer, and the `content image` is the image you want to transfer the style to. \n### The `feature extractor` is the neural network used to extract the features from the images. \n### The number of `style layers` and `content layers` are the number of layers in the feature extractor used to extract the style and content features respectively. \n### The `epochs`, `learning_rate`, `steps_per_epoch`, `style_weight`, `content_weight`, and `total_variation_weight` are all **hyperparameters** that affect the style transfer process. \n### The style weight controls how much the style image affects the output image, the content weight controls how much the content image affects the output image, and the variation weight controls how much the total variation of the output image affects the output image. The total variation of an image is the sum of the absolute differences for neighboring pixel-values in the image. The total variation loss is used to smooth the output image. The higher the variation weight, the smoother the output image will be."""
+
 
 
     outputs = [output_image, style_loss, content_loss, var_loss, curr_epoch, curr_step]
@@ -299,7 +285,10 @@ def main():
         fn=fit_style_transfer,
         inputs=inputs,
         outputs=outputs,
+        title=title,
+        description=description,
         examples=examples,
+        theme='gradio/monochrome' 
         
     )
 

model.py

@@ -6,6 +6,8 @@ from keras import backend as K
 
 class NeuralStyleTransfer:
     def __init__(self, style_image, content_image, extractor, n_style_layers=5, n_content_layers=5):
+
+        self.extractor = extractor
         # load the model
         if extractor == "inception_v3":
             self.feature_extractor = tf.keras.applications.InceptionV3(
@@ -69,31 +71,19 @@ class NeuralStyleTransfer:
 
         return image
 
-    def imshow(self, image, title=None):
-        """displays an image with a corresponding title"""
-        if len(image.shape) > 3:
-            image = tf.squeeze(image, axis=0)
-
-        plt.imshow(image)
-        if title:
-            plt.title(title)
-
-    def show_images_with_objects(self, images, titles=[]):
-        """displays a row of images with corresponding titles"""
-        if len(images) != len(titles):
-            return
-
-        plt.figure(figsize=(20, 12))
-        for idx, (image, title) in enumerate(zip(images, titles)):
-            plt.subplot(1, len(images), idx + 1)
-            plt.xticks([])
-            plt.yticks([])
-            self.imshow(image, title)
 
     def _preprocess_image(self, image):
         image = tf.cast(image, dtype=tf.float32)
-        image = (image / 127.5) - 1.0
-
+        
+        if self.extractor == "inception_v3":
+            image = image / 127.5 - 1
+        elif self.extractor == "vgg19":
+            image = tf.keras.applications.vgg19.preprocess_input(image)
+        elif self.extractor == "resnet50":
+            image = tf.keras.applications.resnet50.preprocess_input(image)
+        elif self.extractor == "mobilenet_v2":
+            # scale pixel between -1 and 1
+            image = image / 127.5 - 1
         return image
 
     def get_output_layers(self):
@@ -219,6 +209,11 @@ class NeuralStyleTransfer:
                 )
             ]
         )
+
+        style_loss /= self.n_style_layers
+        content_loss /= self.n_content_layers
+
+        
         total_loss = style_loss + content_loss
         return total_loss, style_loss, content_loss