Initial commit

ML/TensorFlow/CNN_architectures/LeNet5/README.md

@@ -0,0 +1,5 @@
+[Original Paper - GradientBased Learning Applied to Document Recognition (1998)](http://vision.stanford.edu/cs598_spring07/papers/Lecun98.pdf)  
+[Related Video](https://www.youtube.com/watch?v=fcOW-Zyb5Bo)
+
+Some other useful links
+- [Understanding and Implementing LeNet-5 CNN Architecture](https://towardsdatascience.com/understanding-and-implementing-lenet-5-cnn-architecture-deep-learning-a2d531ebc342)

ML/TensorFlow/CNN_architectures/LeNet5/lenet5.py

@@ -0,0 +1,78 @@
+# Tensorflow v.2.3.1
+
+"""
+Programmed by the-robot <https://github.com/the-robot>
+"""
+
+from tensorflow.keras.layers import (
+    AveragePooling2D,
+    Conv2D,
+    Dense,
+    Flatten,
+    Input,
+)
+from tensorflow.keras import Model
+import tensorflow as tf
+import typing
+
+tf.config.run_functions_eagerly(True)
+
+@tf.function
+def LeNet5(input_shape: typing.Tuple[int], classes: int = 1000) -> Model:
+    """
+    Implementation of the classic LeNet architecture.
+
+    Arguments:
+    input_shape -- shape of the images of the dataset
+    classes     -- integer, number of classes
+
+    Returns:
+    model       -- a Model() instance in Keras
+
+    Note:
+    because I want to keep it original, I used tanh activation instead of ReLU activation.
+    however based on newer papers, the rectified linear unit (ReLU) performed much faster than
+    tanh activation.
+    """
+
+    # convert input shape into tensor
+    X_input = Input(input_shape)
+
+    # layer 1
+    X = Conv2D(
+        filters = 6,
+        kernel_size = (5, 5),
+        strides = (1, 1),
+        activation = "tanh",
+        padding = "valid",
+    )(X_input)
+    X = AveragePooling2D(pool_size = (2, 2), strides = (2, 2), padding = "valid")(X)
+
+    # layer 2
+    X = Conv2D(
+        filters = 16,
+        kernel_size = (5, 5),
+        strides = (1, 1),
+        activation = "tanh",
+        padding = "valid",
+    )(X)
+    X = AveragePooling2D(pool_size = (2, 2), strides = (2, 2), padding = "valid")(X)
+
+    # layer 3
+    X = Conv2D(
+        filters = 120,
+        kernel_size = (5, 5),
+        strides = (1, 1),
+        activation = "tanh",
+        padding = "valid",
+    )(X)
+
+    # layer 4
+    X = Flatten()(X)
+    X = Dense(units = 84, activation = "tanh")(X)
+
+    # layer 5 (classification layer)
+    X = Dense(units = classes, activation = "softmax")(X)
+
+    model = Model(inputs = X_input, outputs = X, name = "LeNet5")
+    return model

ML/TensorFlow/CNN_architectures/LeNet5/test.py

@@ -0,0 +1,9 @@
+# disable tensorflow debugging messages
+import os
+os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
+
+from lenet5 import LeNet5
+
+if __name__ == "__main__":
+    model = LeNet5(input_shape = (32, 32, 1), classes = 10)
+    model.summary()