New Update to code

Author: EteimZ

templates/Image classification_Tensorflow/code-template.py.jinja

@@ -17,6 +17,7 @@ import numpy as np
 import tensorflow as tf
 from tensorflow import keras
 from tensorflow.keras.preprocessing import image_dataset_from_directory
+from tensorflow.keras.preprocessing.image import img_to_array, array_to_img
 {% if data_format == "Image files" %}
 import urllib
 import zipfile
@@ -77,7 +78,7 @@ batch_size = {{ batch_size }}
 num_epochs = {{ num_epochs}}
 
 {# TODO Add Image_Size #}
-img_size = (224,224)
+img_size = (160,160)
 img_shape = img_size + (3,)
 
 # Set up logging.
@@ -103,9 +104,11 @@ def preprocess(data, name):
 
     {% if data_format == "Image files" %}
     # Read image files to tensorflow dataset.
-    dataset = tf.keras.preprocessing.image.ImageDataGenerator(rescale=1./255, rotation_range=20, horizontal_flip=True)
-    loader = dataset.flow_from_directory(data, target_size=img_size)
-    {# TODO: Add more data_augmentation #}
+    dataset = image_dataset_from_directory(data,
+                                           shuffle=(name=="train"),
+                                           image_size=img_size)
+                                          
+
     {% elif data_format == "Numpy arrays" %}
     images, labels = data
 
@@ -118,33 +121,41 @@ def preprocess(data, name):
     # If images are grayscale, convert to RGB by duplicating channels.
     if images.shape[1] == 1:
         images = np.stack((images[:, 0],) * 3, axis=1)
-    )
+    
 
-    {# This code could be improved #}
-    Pil_image = []
+    images = np.rollaxis(images, 1, 4) # Reshape Image to channels_last
 
-    for i  in range(len(data[0])):
-        Pil_image.append(tf.keras.preprocessing.image.array_to_img(images[i],data_format ="channels_first"))
-        Pil_image[i] = Pil_image[i].resize((224,224)
-        Pil_image[i] = tf.keras.preprocessing.image.img_to_array(Pil_image[i])
+    images = np.asarray([img_to_array(array_to_img(im, scale=False).resize((256,256))) for im in images])
 
-    loader = tf.convert_to_tensor(Pil_image)
+    dataset = images, labels
     {% endif %}
-
-    return loader
+    return dataset
 
 train_loader = preprocess(train_data, "train")
 val_loader = preprocess(val_data, "val")
 test_loader = preprocess(test_data, "test")
 
+{{ header("data augmentation") }}
+
+data_augmentation = tf.keras.Sequential([
+  tf.keras.layers.experimental.preprocessing.RandomFlip('horizontal'),
+  tf.keras.layers.experimental.preprocessing.RandomRotation(0.2),
+])
+
 {{ header("Model") }}
 
-model = tf.keras.applications.MobileNetV2(input_shape=img_shape,
+# Create the base model
+base_model = tf.keras.applications.{{model_func}}(input_shape=img_shape,
                                                include_top=True,
-                                               weights='imagenet', classes=1000)
+                                               weights='imagenet')
+
 
+# using the Keras Functional API
 
-model.trainable = True  
+inputs = tf.keras.Input(shape=img_shape)
+x = data_augmentation(inputs)
+x = tf.keras.applications.{{model_func}}.preprocess_input(x)
+model = base_model(x)
 
 model.compile(optimizer = tf.keras.optimizers.{{ optimizer }}(lr={{lr}}),
               loss = "{{ loss }}",
@@ -153,12 +164,14 @@ model.compile(optimizer = tf.keras.optimizers.{{ optimizer }}(lr={{lr}}),
 model.fit(train_loader,
         batch_size={{batch_size}},
         epochs={{num_epochs}},
-        validation_data=val_loader
+        validation_data=val_loader,
         {% if visualization_tool == "Tensorboard" and checkpoint%}
-        ,callbacks = [tensorboard_callback, checkpoint_dir]
+        callbacks = [tensorboard_callback, checkpoint_dir],
         {% elif checkpoint %}
-        ,callbacks = [checkpoint_dir]
+        callbacks = [checkpoint_dir],
         {% elif visualization_tool == "Tensorboard" %}
-        ,callbacks = [tensorboard_callback]
+        callbacks = [tensorboard_callback],
         {% endif %}
         )
+
+model.evaluate(val_loader)

templates/Image classification_Tensorflow/sidebar.py

@@ -6,24 +6,19 @@
 # option 1: model -> code
 # option 2 – if model has multiple variants: model -> model variant -> code
 MODELS = {
-    "AlexNet": "alexnet",  # single model variant
+    #TODO: Add more models
+    "Xception": "Xception", # single model variant
     "ResNet": {  # multiple model variants
-        "ResNet 18": "resnet18",
-        "ResNet 34": "resnet34",
-        "ResNet 50": "resnet50",
-        "ResNet 101": "resnet101",
-        "ResNet 152": "resnet152",
+        "ResNet 50": "ResNet50",
+        "ResNet 101": "ResNet101",
+        "ResNet 152": "ResNet152",
+        "ResNet 50v2": "ResNet50V2",
+        "ResNet 101v2": "ResNet101V2",
+        "ResNet 152v2": "ResNet152V2",
     },
-    "DenseNet": "densenet",
     "VGG": {
-        "VGG11": "vgg11",
-        "VGG11 with batch normalization": "vgg11_bn",
-        "VGG13": "vgg13",
-        "VGG13 with batch normalization": "vgg13_bn",
-        "VGG16": "vgg16",
-        "VGG16 with batch normalization": "vgg16_bn",
+        "VGG16": "VGG16",
         "VGG19": "vgg19",
-        "VGG19 with batch normalization": "vgg19_bn",
     },
 }