Areg

requirements.txt

@@ -0,0 +1 @@
+tensorflow.__version__ = 2.11.0

run_pipeline.py

@@ -0,0 +1,17 @@
+from Preporcessor import Preprocessor
+import tensorflow as tf
+import argparse
+parser = argparse.ArgumentParser()
+### example path~~~\Tuned_Inseption
+parser.add_argument("--data_path", type=str, help="path to testing images")
+### example path~~~\data\images
+parser.add_argument("--model_path", type=str,  help="path to Stored Trained_Model")
+args = parser.parse_args()
+
+Pretrained_Model = tf.keras.models.load_model(args.model_path)
+pepr_= Preprocessor(Path=args.data_path,Mode="Test")
+Test_X,Test_Y = pepr_.fit_transform()
+Scores =Pretrained_Model.evaluate(Test_X,Test_Y)
+
+print(f"MODEL LOSS==>> {Scores[0]}")
+print(f"MODEL ACCURACY==>> {Scores[1]}")

src/Preprocessor.py

@@ -0,0 +1,53 @@
+import tensorflow as tf
+import pathlib
+class Preprocessor:
+
+    def __init__(self,Path, batch_size=64, img_height=299, img_width=299, validation_rate=0.15, Mode="Train",):
+        """
+        :param Path: Directory where your images are located
+        :param batch_size: 
+        :param img_height: 
+        :param img_width: 
+        :param validation_rate: 
+        :param Mode: if you want to train model and preporcess your train and validation data Initialize in Train Mode if you wont to prepare data for Test, so in test mode
+        """
+        self.batch_size = batch_size
+        self.img_height = img_height
+        self.img_width = img_width
+        self.validation_rate = validation_rate
+        self.Mode = Mode
+        self.Path = pathlib.Path(Path)
+
+    def fit_transform(self):
+        """
+        funtion return resized and labeled dataset form images Directory
+        if function call is in Training mode function return training and validation dataset and if in test mode return data for testing
+        :return:
+        """
+        if self.Mode == "Train":
+            train_data_set = tf.keras.utils.image_dataset_from_directory(
+                self.Path,
+                validation_split=self.validation_rate,
+                subset="training",
+                seed=111,
+                image_size=(self.img_height, self.img_width),
+                batch_size=self.batch_size,
+                shuffle=True)
+            validation_data_set = tf.keras.utils.image_dataset_from_directory(
+                self.Path,
+                validation_split=self.validation_rate,
+                subset="validation",
+                seed=111,
+                image_size=(self.img_height, self.img_width),
+                batch_size=self.batch_size,
+                shuffle=True)
+            return train_data_set,validation_data_set
+        else:
+            Test_data_set = tf.keras.utils.image_dataset_from_directory(
+                self.Path,
+                image_size=(self.img_height, self.img_width),
+                batch_size=320,
+                shuffle=False)
+            x=list(Test_data_set)[0][0]
+            y=list(Test_data_set)[0][1]
+            return x,y

src/model.py

@@ -0,0 +1,29 @@
+import tensorflow as tf
+from Preporcessor import Preprocessor
+Prepr_ = Preprocessor(Path="Your train_data_path ",Mode="Train")
+## Loading and spltting data into train and validation sets
+training_data, validation_data = Prepr_.fit_transform()
+### We use GoogleNet/IncpetionV3 as our base model 
+
+
+Base_model = tf.keras.applications.InceptionV3(include_top=False,weights="imagenet")
+Base_model.trainable=False
+model = tf.keras.Sequential()
+model.add(tf.keras.layers.Rescaling(scale=1./127.5, offset=-1))
+model.add(Base_model)
+model.add(tf.keras.layers.GlobalAvgPool2D())
+## After Inception pre_ptained block we add 3 Danse layers and one Dropout layer and alos chane output layer into 10 layers Danse with softmax
+## trainable patamets are ~~58000
+model.add(tf.keras.layers.Dense(28,activation="relu"))
+model.add(tf.keras.layers.Dense(18,activation="relu"))
+model.add(tf.keras.layers.Dropout(rate=0.2))
+model.add(tf.keras.layers.Dense(16,activation="relu"))
+model.add(tf.keras.layers.Dense(10,activation="softmax"))
+model.build(input_shape=(64,299,299,3))
+optimizer = tf.keras.optimizers.Adam()
+model.compile(optimizer=optimizer,loss="sparse_categorical_crossentropy",metrics=["accuracy"])
+
+print(model.summary())
+
+model.fit(training_data,validation_data=validation_data,epochs=10)
+