This is a simple example of creating sequential convolutional neural network model in tensorflow keras and training it to recognize MNIST digits dataset that’s today’s “Hello World” of image recognition.

Import required libraries

from tensorflow.keras import layers
from tensorflow.keras import models

Create Sequential model:

model = models.Sequential()
model.add(layers.Conv2D(32, (3,3), activation='relu', input_shape=(28,28,1)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3,3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3,3), activation='relu'))
model.add(layers.Flatten())
model.add(layers.Dense(64, activation='relu'))
model.add(layers.Dense(10, activation='softmax'))

Print out summary of the model:

model.summary()

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d (Conv2D)              (None, 26, 26, 32)        320       
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 13, 13, 32)        0         
_________________________________________________________________
conv2d_1 (Conv2D)            (None, 11, 11, 64)        18496     
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 5, 5, 64)          0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 3, 3, 64)          36928     
_________________________________________________________________
flatten (Flatten)            (None, 576)               0         
_________________________________________________________________
dense (Dense)                (None, 64)                36928     
_________________________________________________________________
dense_1 (Dense)              (None, 10)                650       
=================================================================
Total params: 93,322
Trainable params: 93,322
Non-trainable params: 0
_________________________________________________________________

Import MNIST dataset and to_categorical function

from tensorflow.keras.datasets import mnist
from tensorflow.keras.utils import to_categorical

Init train and test data

(train_images, train_labels), (test_images, test_labels) = mnist.load_data()

Reshape and normalize data and labels

train_images = train_images.reshape((60000, 28, 28, 1))
train_images = train_images.astype('float32') / 255
test_images = test_images.reshape((10000, 28, 28, 1))
test_images = test_images.astype('float32') / 255
train_labels = to_categorical(train_labels)
test_labels = to_categorical(test_labels)

Train the model

model.compile(optimizer='rmsprop', loss='categorical_crossentropy',
             metrics=['accuracy'])
model.fit(train_images, train_labels, epochs=5, batch_size=64)

Epoch 1/5
60000/60000 [==============================] - 5s 77us/step - loss: 0.1711 - acc: 0.9471
Epoch 2/5
60000/60000 [==============================] - 4s 59us/step - loss: 0.0488 - acc: 0.9849
Epoch 3/5
60000/60000 [==============================] - 4s 60us/step - loss: 0.0336 - acc: 0.9898
Epoch 4/5
60000/60000 [==============================] - 4s 59us/step - loss: 0.0255 - acc: 0.9924
Epoch 5/5
60000/60000 [==============================] - 4s 59us/step - loss: 0.0196 - acc: 0.9940

<tensorflow.python.keras.callbacks.History at 0x7f07860745f8>

Save trained model

model.save('mnist_cnn.h5')

Load and evaluate the model (test the accuracy on test data)

from tensorflow.keras.models import load_model

model = load_model('mnist_cnn.h5')
test_loss, test_acc = model.evaluate(test_images, test_labels)

10000/10000 [==============================] - 0s 49us/step

Print out the accuracy

test_acc

0.9918

Show example test image:

import matplotlib.pyplot as plt

test_img = test_images[0].reshape(28,28)
plt.imshow(test_img, cmap='Greys')

<matplotlib.image.AxesImage at 0x7f076011e5f8>

png

Predict example test image using the model

prediction_probs = model.predict(test_img.reshape(1, 28,28,1)).flatten()

prediction_probs.argmax()

7

This experiment was made after reading Deep Learning with Python by Francois Chollet. I strongly recomment this book to everyone interested in python, deep learning, tensorflow and keras.

Deep Learning with Python

Source code is available to download at my github repository:

https://github.com/sikora507/keras_mnist