Image Classification

In this project, you'll classify images from the CIFAR-10 dataset. The dataset consists of airplanes, dogs, cats, and other objects. You'll preprocess the images, then train a convolutional neural network on all the samples. The images need to be normalized and the labels need to be one-hot encoded. You'll get to apply what you learned and build a convolutional, max pooling, dropout, and fully connected layers. At the end, you'll get to see your neural network's predictions on the sample images.

Get the Data

Run the following cell to download the CIFAR-10 dataset for python.


In [1]:
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
from urllib.request import urlretrieve
from os.path import isfile, isdir
from tqdm import tqdm
import problem_unittests as tests
import tarfile

cifar10_dataset_folder_path = 'cifar-10-batches-py'

class DLProgress(tqdm):
    last_block = 0

    def hook(self, block_num=1, block_size=1, total_size=None):
        self.total = total_size
        self.update((block_num - self.last_block) * block_size)
        self.last_block = block_num

if not isfile('cifar-10-python.tar.gz'):
    with DLProgress(unit='B', unit_scale=True, miniters=1, desc='CIFAR-10 Dataset') as pbar:
        urlretrieve(
            'https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz',
            'cifar-10-python.tar.gz',
            pbar.hook)

if not isdir(cifar10_dataset_folder_path):
    with tarfile.open('cifar-10-python.tar.gz') as tar:
        tar.extractall()
        tar.close()


tests.test_folder_path(cifar10_dataset_folder_path)


All files found!

Explore the Data

The dataset is broken into batches to prevent your machine from running out of memory. The CIFAR-10 dataset consists of 5 batches, named data_batch_1, data_batch_2, etc.. Each batch contains the labels and images that are one of the following:

  • airplane
  • automobile
  • bird
  • cat
  • deer
  • dog
  • frog
  • horse
  • ship
  • truck

Understanding a dataset is part of making predictions on the data. Play around with the code cell below by changing the batch_id and sample_id. The batch_id is the id for a batch (1-5). The sample_id is the id for a image and label pair in the batch.

Ask yourself "What are all possible labels?", "What is the range of values for the image data?", "Are the labels in order or random?". Answers to questions like these will help you preprocess the data and end up with better predictions.


In [2]:
%matplotlib inline
%config InlineBackend.figure_format = 'retina'

import helper
import numpy as np

# Explore the dataset
batch_id = 1
sample_id = 5
helper.display_stats(cifar10_dataset_folder_path, batch_id, sample_id)


Stats of batch 1:
Samples: 10000
Label Counts: {0: 1005, 1: 974, 2: 1032, 3: 1016, 4: 999, 5: 937, 6: 1030, 7: 1001, 8: 1025, 9: 981}
First 20 Labels: [6, 9, 9, 4, 1, 1, 2, 7, 8, 3, 4, 7, 7, 2, 9, 9, 9, 3, 2, 6]

Example of Image 5:
Image - Min Value: 0 Max Value: 252
Image - Shape: (32, 32, 3)
Label - Label Id: 1 Name: automobile

Implement Preprocess Functions

Normalize

In the cell below, implement the normalize function to take in image data, x, and return it as a normalized Numpy array. The values should be in the range of 0 to 1, inclusive. The return object should be the same shape as x.


In [3]:
def normalize(x):
    """
    Normalize a list of sample image data in the range of 0 to 1
    : x: List of image data.  The image shape is (32, 32, 3)
    : return: Numpy array of normalize data
    """
    # TODO: Implement Function
    # Normalize RGB for each image data
    norm_image_data = np.ndarray(x.shape, np.float32)
    for i, image_data in enumerate(x):
        temp_image_data = np.ndarray(x.shape, np.float32)
        temp_image_data = image_data
        temp_image_data[:,:,0] = abs((temp_image_data[:,:,0] - 128)/128)
        temp_image_data[:,:,1] = abs((temp_image_data[:,:,1] - 128)/128)
        temp_image_data[:,:,2] = abs((temp_image_data[:,:,2] - 128)/128)
        norm_image_data[i] = temp_image_data
    #print('Sample normalized image data: {}'.format(norm_image_data[2].max()))
    return norm_image_data


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_normalize(normalize)


Tests Passed

One-hot encode

Just like the previous code cell, you'll be implementing a function for preprocessing. This time, you'll implement the one_hot_encode function. The input, x, are a list of labels. Implement the function to return the list of labels as One-Hot encoded Numpy array. The possible values for labels are 0 to 9. The one-hot encoding function should return the same encoding for each value between each call to one_hot_encode. Make sure to save the map of encodings outside the function.

Hint: Don't reinvent the wheel.


In [4]:
def one_hot_encode(x):
    """
    One hot encode a list of sample labels. Return a one-hot encoded vector for each label.
    : x: List of sample Labels
    : return: Numpy array of one-hot encoded labels
    """
    # TODO: Implement Function
    one_hot = np.zeros(shape=[len(x), 10])
    for i, label_id in enumerate(x):
        one_hot[i, label_id] = True
    return one_hot

one_hot_encoding_map = {0:'airplane', 1:'automobile', 2:'bird', 3:'cat', 4:'deer', 5:'dog', 6:'frog', 7:'horse',\
               8:'ship', 9:'truck'}
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_one_hot_encode(one_hot_encode)


Tests Passed

Randomize Data

As you saw from exploring the data above, the order of the samples are randomized. It doesn't hurt to randomize it again, but you don't need to for this dataset.

Preprocess all the data and save it

Running the code cell below will preprocess all the CIFAR-10 data and save it to file. The code below also uses 10% of the training data for validation.


In [16]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
# Preprocess Training, Validation, and Testing Data
helper.preprocess_and_save_data(cifar10_dataset_folder_path, normalize, one_hot_encode)


Sample normalized image data: 1.0
Sample normalized image data: 1.0
Sample normalized image data: 1.0
Sample normalized image data: 1.0
Sample normalized image data: 1.0
Sample normalized image data: 1.0
Sample normalized image data: 1.0

Check Point

This is your first checkpoint. If you ever decide to come back to this notebook or have to restart the notebook, you can start from here. The preprocessed data has been saved to disk.


In [1]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import pickle
import problem_unittests as tests
import helper

# Load the Preprocessed Validation data
valid_features, valid_labels = pickle.load(open('preprocess_validation.p', mode='rb'))

Build the network

For the neural network, you'll build each layer into a function. Most of the code you've seen has been outside of functions. To test your code more thoroughly, we require that you put each layer in a function. This allows us to give you better feedback and test for simple mistakes using our unittests before you submit your project.

If you're finding it hard to dedicate enough time for this course a week, we've provided a small shortcut to this part of the project. In the next couple of problems, you'll have the option to use TensorFlow Layers or TensorFlow Layers (contrib) to build each layer, except "Convolutional & Max Pooling" layer. TF Layers is similar to Keras's and TFLearn's abstraction to layers, so it's easy to pickup.

If you would like to get the most of this course, try to solve all the problems without TF Layers. Let's begin!

Input

The neural network needs to read the image data, one-hot encoded labels, and dropout keep probability. Implement the following functions

  • Implement neural_net_image_input
    • Return a TF Placeholder
    • Set the shape using image_shape with batch size set to None.
    • Name the TensorFlow placeholder "x" using the TensorFlow name parameter in the TF Placeholder.
  • Implement neural_net_label_input
    • Return a TF Placeholder
    • Set the shape using n_classes with batch size set to None.
    • Name the TensorFlow placeholder "y" using the TensorFlow name parameter in the TF Placeholder.
  • Implement neural_net_keep_prob_input
    • Return a TF Placeholder for dropout keep probability.
    • Name the TensorFlow placeholder "keep_prob" using the TensorFlow name parameter in the TF Placeholder.

These names will be used at the end of the project to load your saved model.

Note: None for shapes in TensorFlow allow for a dynamic size.


In [2]:
import tensorflow as tf

def neural_net_image_input(image_shape):
    """
    Return a Tensor for a bach of image input
    : image_shape: Shape of the images
    : return: Tensor for image input.
    """
    # TODO: Implement Function
    image_input = tf.placeholder(tf.float32, shape=(None, image_shape[0], image_shape[1], image_shape[2]), name='x')
    return image_input


def neural_net_label_input(n_classes):
    """
    Return a Tensor for a batch of label input
    : n_classes: Number of classes
    : return: Tensor for label input.
    """
    # TODO: Implement Function
    label_input = tf.placeholder(tf.float32, shape=(None, n_classes), name='y')
    return label_input


def neural_net_keep_prob_input():
    """
    Return a Tensor for keep probability
    : return: Tensor for keep probability.
    """
    # TODO: Implement Function
    keep_prob_input = tf.placeholder(tf.float32, name='keep_prob')
    return keep_prob_input


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tf.reset_default_graph()
tests.test_nn_image_inputs(neural_net_image_input)
tests.test_nn_label_inputs(neural_net_label_input)
tests.test_nn_keep_prob_inputs(neural_net_keep_prob_input)


Image Input Tests Passed.
Label Input Tests Passed.
Keep Prob Tests Passed.

Convolution and Max Pooling Layer

Convolution layers have a lot of success with images. For this code cell, you should implement the function conv2d_maxpool to apply convolution then max pooling:

  • Create the weight and bias using conv_ksize, conv_num_outputs and the shape of x_tensor.
  • Apply a convolution to x_tensor using weight and conv_strides.
    • We recommend you use same padding, but you're welcome to use any padding.
  • Add bias
  • Add a nonlinear activation to the convolution.
  • Apply Max Pooling using pool_ksize and pool_strides.
    • We recommend you use same padding, but you're welcome to use any padding.

Note: You can't use TensorFlow Layers or TensorFlow Layers (contrib) for this layer. You're free to use any TensorFlow package for all the other layers.


In [3]:
def conv2d_maxpool(x_tensor, conv_num_outputs, conv_ksize, conv_strides, pool_ksize, pool_strides):
    """
    Apply convolution then max pooling to x_tensor
    :param x_tensor: TensorFlow Tensor
    :param conv_num_outputs: Number of outputs for the convolutional layer
    :param conv_strides: Stride 2-D Tuple for convolution
    :param pool_ksize: kernal size 2-D Tuple for pool
    :param pool_strides: Stride 2-D Tuple for pool
    : return: A tensor that represents convolution and max pooling of x_tensor
    """
    # TODO: Implement Function
    print('Conv_ksize: ', conv_ksize, ' Conv_strides: ', conv_strides, ' Conv output depth:', conv_num_outputs, \
          x_tensor.get_shape().as_list(), ' Pool ksize: ', pool_ksize, ' Pool strides: ', pool_strides)
    #Convolution and max pool Parameters
    input_depth = x_tensor.get_shape().as_list()[3]
    output_depth = conv_num_outputs
    weight = tf.Variable(tf.truncated_normal([conv_ksize[0], conv_ksize[1], input_depth, output_depth], mean=0.0, stddev=0.1))
    biases = tf.Variable(tf.truncated_normal(output_depth))
    strides = [1, conv_strides[0], conv_strides[1], 1]
    pool_strides = [1, pool_strides[0], pool_strides[1], 1]
    
    #Convolution & Max pool
    conv2d_1 = tf.nn.conv2d(x_tensor, weight, strides, padding='SAME')
    conv2d_1 = tf.nn.bias_add(conv2d_1, biases)
    conv2d_1 = tf.nn.relu(conv2d_1)
    conv2d_1 = tf.nn.max_pool(conv2d_1, [1, pool_ksize[0], pool_ksize[1], 1], pool_strides, padding='SAME')
    return conv2d_1


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_con_pool(conv2d_maxpool)


Conv_ksize:  (2, 2)  Conv_strides:  (4, 4)  Conv output depth: 10 [None, 32, 32, 5]  Pool ksize:  (2, 2)  Pool strides:  (2, 2)
Tests Passed

Flatten Layer

Implement the flatten function to change the dimension of x_tensor from a 4-D tensor to a 2-D tensor. The output should be the shape (Batch Size, Flattened Image Size). You can use TensorFlow Layers or TensorFlow Layers (contrib) for this layer.


In [4]:
def flatten(x_tensor):
    """
    Flatten x_tensor to (Batch Size, Flattened Image Size)
    : x_tensor: A tensor of size (Batch Size, ...), where ... are the image dimensions.
    : return: A tensor of size (Batch Size, Flattened Image Size).
    """
    # TODO: Implement Function
    #print(x_tensor.get_shape().as_list()[3])
    #print(x_tensor.get_shape().as_list())
    h = x_tensor.get_shape().as_list()[1]
    w = x_tensor.get_shape().as_list()[2]
    d = x_tensor.get_shape().as_list()[3]
    flattened_tensor = tf.reshape(x_tensor, [-1, h*w*d])
    #print(flattened_tensor.get_shape().as_list())
    return flattened_tensor


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_flatten(flatten)


Tests Passed

Fully-Connected Layer

Implement the fully_conn function to apply a fully connected layer to x_tensor with the shape (Batch Size, num_outputs). You can use TensorFlow Layers or TensorFlow Layers (contrib) for this layer.


In [5]:
def fully_conn(x_tensor, num_outputs):
    """
    Apply a fully connected layer to x_tensor using weight and bias
    : x_tensor: A 2-D tensor where the first dimension is batch size.
    : num_outputs: The number of output that the new tensor should be.
    : return: A 2-D tensor where the second dimension is num_outputs.
    """
    # TODO: Implement Function
    weight_rows = x_tensor.get_shape().as_list()[1]
    weight = tf.Variable(tf.truncated_normal([weight_rows, num_outputs], mean=0.0, stddev=0.1))
    biases = tf.Variable(tf.truncated_normal([num_outputs]))
    fc1 = tf.add(tf.matmul(x_tensor, weight), biases)
    fc1 = tf.nn.relu(fc1)
    return fc1


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_fully_conn(fully_conn)


Tests Passed

Output Layer

Implement the output function to apply a fully connected layer to x_tensor with the shape (Batch Size, num_outputs). You can use TensorFlow Layers or TensorFlow Layers (contrib) for this layer.

Note: Activation, softmax, or cross entropy shouldn't be applied to this.


In [6]:
def output(x_tensor, num_outputs):
    """
    Apply a output layer to x_tensor using weight and bias
    : x_tensor: A 2-D tensor where the first dimension is batch size.
    : num_outputs: The number of output that the new tensor should be.
    : return: A 2-D tensor where the second dimension is num_outputs.
    """
    # TODO: Implement Function
    weight = tf.Variable(tf.truncated_normal([x_tensor.get_shape().as_list()[1], num_outputs], mean=0.0, stddev=0.1))
    biases = tf.Variable(tf.zeros([num_outputs]))
    out = tf.add(tf.matmul(x_tensor, weight), biases)
    return out


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_output(output)


Tests Passed

Create Convolutional Model

Implement the function conv_net to create a convolutional neural network model. The function takes in a batch of images, x, and outputs logits. Use the layers you created above to create this model:

  • Apply 1, 2, or 3 Convolution and Max Pool layers
  • Apply a Flatten Layer
  • Apply 1, 2, or 3 Fully Connected Layers
  • Apply an Output Layer
  • Return the output
  • Apply TensorFlow's Dropout to one or more layers in the model using keep_prob.

In [7]:
def conv_net(x, keep_prob):
    """
    Create a convolutional neural network model
    : x: Placeholder tensor that holds image data.
    : keep_prob: Placeholder tensor that hold dropout keep probability.
    : return: Tensor that represents logits
    """
    # TODO: Apply 1, 2, or 3 Convolution and Max Pool layers
    #    Play around with different number of outputs, kernel size and stride
    # Function Definition from Above:
    #    conv2d_maxpool(x_tensor, conv_num_outputs, conv_ksize, conv_strides, pool_ksize, pool_strides)
    conv2d_1 = conv2d_maxpool(x, 10, (5, 5), (1, 1), (2, 2), (2, 2))
    conv2d_2 = conv2d_maxpool(conv2d_1, 32, (5, 5), (1, 1), (2, 2), (2, 2))
    conv2d_3 = conv2d_maxpool(conv2d_2, 64, (5, 5), (1, 1), (2, 2), (2, 2))
    

    # TODO: Apply a Flatten Layer
    # Function Definition from Above:
    #   flatten(x_tensor)
    flattened_tensor = flatten(conv2d_3)
    

    # TODO: Apply 1, 2, or 3 Fully Connected Layers
    #    Play around with different number of outputs
    # Function Definition from Above:
    #   fully_conn(x_tensor, num_outputs)
    fc1 = fully_conn(flattened_tensor, 64)
    fc1 = tf.nn.dropout(fc1, keep_prob)
    fc2 = fully_conn(fc1, 32)
    fc2 = tf.nn.dropout(fc2, keep_prob)
    
    # TODO: Apply an Output Layer
    #    Set this to the number of classes
    # Function Definition from Above:
    #   output(x_tensor, num_outputs)
    logits = output(fc2, 10)
    
    # TODO: return output
    return logits


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""

##############################
## Build the Neural Network ##
##############################

# Remove previous weights, bias, inputs, etc..
tf.reset_default_graph()

# Inputs
x = neural_net_image_input((32, 32, 3))
y = neural_net_label_input(10)
keep_prob = neural_net_keep_prob_input()

# Model
logits = conv_net(x, keep_prob)

# Name logits Tensor, so that is can be loaded from disk after training
logits = tf.identity(logits, name='logits')

# Loss and Optimizer
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits, labels=y))
optimizer = tf.train.AdamOptimizer().minimize(cost)

# Accuracy
correct_pred = tf.equal(tf.argmax(logits, 1), tf.argmax(y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32), name='accuracy')

tests.test_conv_net(conv_net)


Conv_ksize:  (5, 5)  Conv_strides:  (1, 1)  Conv output depth: 10 [None, 32, 32, 3]  Pool ksize:  (2, 2)  Pool strides:  (2, 2)
Conv_ksize:  (5, 5)  Conv_strides:  (1, 1)  Conv output depth: 32 [None, 16, 16, 10]  Pool ksize:  (2, 2)  Pool strides:  (2, 2)
Conv_ksize:  (5, 5)  Conv_strides:  (1, 1)  Conv output depth: 64 [None, 8, 8, 32]  Pool ksize:  (2, 2)  Pool strides:  (2, 2)
Conv_ksize:  (5, 5)  Conv_strides:  (1, 1)  Conv output depth: 10 [None, 32, 32, 3]  Pool ksize:  (2, 2)  Pool strides:  (2, 2)
Conv_ksize:  (5, 5)  Conv_strides:  (1, 1)  Conv output depth: 32 [None, 16, 16, 10]  Pool ksize:  (2, 2)  Pool strides:  (2, 2)
Conv_ksize:  (5, 5)  Conv_strides:  (1, 1)  Conv output depth: 64 [None, 8, 8, 32]  Pool ksize:  (2, 2)  Pool strides:  (2, 2)
Neural Network Built!

Train the Neural Network

Single Optimization

Implement the function train_neural_network to do a single optimization. The optimization should use optimizer to optimize in session with a feed_dict of the following:

  • x for image input
  • y for labels
  • keep_prob for keep probability for dropout

This function will be called for each batch, so tf.global_variables_initializer() has already been called.

Note: Nothing needs to be returned. This function is only optimizing the neural network.


In [8]:
def train_neural_network(session, optimizer, keep_probability, feature_batch, label_batch):
    """
    Optimize the session on a batch of images and labels
    : session: Current TensorFlow session
    : optimizer: TensorFlow optimizer function
    : keep_probability: keep probability
    : feature_batch: Batch of Numpy image data
    : label_batch: Batch of Numpy label data
    """
    # TODO: Implement Function
    session.run(optimizer, feed_dict={ x: feature_batch, y: label_batch, keep_prob: keep_probability})


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_train_nn(train_neural_network)


Tests Passed

Show Stats

Implement the function print_stats to print loss and validation accuracy. Use the global variables valid_features and valid_labels to calculate validation accuracy. Use a keep probability of 1.0 to calculate the loss and validation accuracy.


In [9]:
def print_stats(session, feature_batch, label_batch, cost, accuracy):
    """
    Print information about loss and validation accuracy
    : session: Current TensorFlow session
    : feature_batch: Batch of Numpy image data
    : label_batch: Batch of Numpy label data
    : cost: TensorFlow cost function
    : accuracy: TensorFlow accuracy function
    """
    # TODO: Implement Function
    loss = session.run(cost, feed_dict={x: feature_batch, y: label_batch, keep_prob: 1.0})
    valid_accuracy = session.run(accuracy, feed_dict={x: valid_features, y: valid_labels, keep_prob: 1.0})
    print('Cost: ', loss)
    print('Accuracy: ', valid_accuracy)

Hyperparameters

Tune the following parameters:

  • Set epochs to the number of iterations until the network stops learning or start overfitting
  • Set batch_size to the highest number that your machine has memory for. Most people set them to common sizes of memory:
    • 64
    • 128
    • 256
    • ...
  • Set keep_probability to the probability of keeping a node using dropout

In [10]:
# TODO: Tune Parameters
epochs = 20
batch_size = 64
keep_probability = 0.5

Train on a Single CIFAR-10 Batch

Instead of training the neural network on all the CIFAR-10 batches of data, let's use a single batch. This should save time while you iterate on the model to get a better accuracy. Once the final validation accuracy is 50% or greater, run the model on all the data in the next section.


In [11]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
print('Checking the Training on a Single Batch...')
with tf.Session() as sess:
    # Initializing the variables
    sess.run(tf.global_variables_initializer())
    
    # Training cycle
    for epoch in range(epochs):
        batch_i = 1
        for batch_features, batch_labels in helper.load_preprocess_training_batch(batch_i, batch_size):
            train_neural_network(sess, optimizer, keep_probability, batch_features, batch_labels)
        print('Epoch {:>2}, CIFAR-10 Batch {}:  '.format(epoch + 1, batch_i), end='')
        print_stats(sess, batch_features, batch_labels, cost, accuracy)


Checking the Training on a Single Batch...
Epoch  1, CIFAR-10 Batch 1:  Cost:  2.13784
Accuracy:  0.2276
Epoch  2, CIFAR-10 Batch 1:  Cost:  2.03116
Accuracy:  0.2682
Epoch  3, CIFAR-10 Batch 1:  Cost:  2.00004
Accuracy:  0.3094
Epoch  4, CIFAR-10 Batch 1:  Cost:  1.91627
Accuracy:  0.33
Epoch  5, CIFAR-10 Batch 1:  Cost:  1.87016
Accuracy:  0.3306
Epoch  6, CIFAR-10 Batch 1:  Cost:  1.75752
Accuracy:  0.3698
Epoch  7, CIFAR-10 Batch 1:  Cost:  1.77981
Accuracy:  0.3654
Epoch  8, CIFAR-10 Batch 1:  Cost:  1.61618
Accuracy:  0.3902
Epoch  9, CIFAR-10 Batch 1:  Cost:  1.54356
Accuracy:  0.3952
Epoch 10, CIFAR-10 Batch 1:  Cost:  1.48269
Accuracy:  0.4032
Epoch 11, CIFAR-10 Batch 1:  Cost:  1.38914
Accuracy:  0.41
Epoch 12, CIFAR-10 Batch 1:  Cost:  1.27977
Accuracy:  0.4228
Epoch 13, CIFAR-10 Batch 1:  Cost:  1.15595
Accuracy:  0.4352
Epoch 14, CIFAR-10 Batch 1:  Cost:  1.12182
Accuracy:  0.4272
Epoch 15, CIFAR-10 Batch 1:  Cost:  1.02883
Accuracy:  0.4362
Epoch 16, CIFAR-10 Batch 1:  Cost:  1.02598
Accuracy:  0.428
Epoch 17, CIFAR-10 Batch 1:  Cost:  0.971006
Accuracy:  0.4364
Epoch 18, CIFAR-10 Batch 1:  Cost:  0.896097
Accuracy:  0.4246
Epoch 19, CIFAR-10 Batch 1:  Cost:  0.903668
Accuracy:  0.4158
Epoch 20, CIFAR-10 Batch 1:  Cost:  0.893489
Accuracy:  0.4376

Fully Train the Model

Now that you got a good accuracy with a single CIFAR-10 batch, try it with all five batches.


In [14]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
save_model_path = './image_classification'

print('Training...')
with tf.Session() as sess:
    # Initializing the variables
    sess.run(tf.global_variables_initializer())
    
    # Training cycle
    for epoch in range(epochs):
        # Loop over all batches
        n_batches = 5
        for batch_i in range(1, n_batches + 1):
            for batch_features, batch_labels in helper.load_preprocess_training_batch(batch_i, batch_size):
                train_neural_network(sess, optimizer, keep_probability, batch_features, batch_labels)
            print('Epoch {:>2}, CIFAR-10 Batch {}:  '.format(epoch + 1, batch_i), end='')
            print_stats(sess, batch_features, batch_labels, cost, accuracy)
            
    # Save Model
    saver = tf.train.Saver()
    save_path = saver.save(sess, save_model_path)


Training...
Epoch  1, CIFAR-10 Batch 1:  110.713
0.1076
Epoch  1, CIFAR-10 Batch 2:  74.3361
0.1096
Epoch  1, CIFAR-10 Batch 3:  5.52944
0.071
Epoch  1, CIFAR-10 Batch 4:  9.36288
0.0754
Epoch  1, CIFAR-10 Batch 5:  4.26181
0.1098
Epoch  2, CIFAR-10 Batch 1:  15.4357
0.1084
Epoch  2, CIFAR-10 Batch 2:  24.4925
0.0862
Epoch  2, CIFAR-10 Batch 3:  2.39048
0.0866
Epoch  2, CIFAR-10 Batch 4:  2.31181
0.105
Epoch  2, CIFAR-10 Batch 5:  2.40985
0.104
Epoch  3, CIFAR-10 Batch 1:  4.8725
0.1054
Epoch  3, CIFAR-10 Batch 2:  14.5355
0.1048
Epoch  3, CIFAR-10 Batch 3:  2.31345
0.104
Epoch  3, CIFAR-10 Batch 4:  2.25305
0.1032
Epoch  3, CIFAR-10 Batch 5:  2.3953
0.1038
Epoch  4, CIFAR-10 Batch 1:  2.38861
0.1042
Epoch  4, CIFAR-10 Batch 2:  11.6032
0.1044
Epoch  4, CIFAR-10 Batch 3:  2.30092
0.1048
Epoch  4, CIFAR-10 Batch 4:  2.26616
0.1044
Epoch  4, CIFAR-10 Batch 5:  2.29977
0.1044
Epoch  5, CIFAR-10 Batch 1:  2.38343
0.1042
Epoch  5, CIFAR-10 Batch 2:  8.44051
0.1042
Epoch  5, CIFAR-10 Batch 3:  2.2997
0.1042
Epoch  5, CIFAR-10 Batch 4:  2.3014
0.1044
Epoch  5, CIFAR-10 Batch 5:  2.30125
0.1036
Epoch  6, CIFAR-10 Batch 1:  2.29232
0.1032
Epoch  6, CIFAR-10 Batch 2:  6.25544
0.1034
Epoch  6, CIFAR-10 Batch 3:  2.3005
0.1038
Epoch  6, CIFAR-10 Batch 4:  2.30176
0.0954
Epoch  6, CIFAR-10 Batch 5:  2.30184
0.0954
Epoch  7, CIFAR-10 Batch 1:  2.29496
0.1032
Epoch  7, CIFAR-10 Batch 2:  4.19178
0.103
Epoch  7, CIFAR-10 Batch 3:  2.30065
0.103
Epoch  7, CIFAR-10 Batch 4:  2.3015
0.0948
Epoch  7, CIFAR-10 Batch 5:  2.30183
0.0952
Epoch  8, CIFAR-10 Batch 1:  2.30263
0.1028
Epoch  8, CIFAR-10 Batch 2:  3.17542
0.1006
Epoch  8, CIFAR-10 Batch 3:  2.30087
0.1006
Epoch  8, CIFAR-10 Batch 4:  2.30132
0.0944
Epoch  8, CIFAR-10 Batch 5:  2.30187
0.0944
Epoch  9, CIFAR-10 Batch 1:  2.30276
0.1024
Epoch  9, CIFAR-10 Batch 2:  2.30286
0.1006
Epoch  9, CIFAR-10 Batch 3:  2.30076
0.1002
Epoch  9, CIFAR-10 Batch 4:  2.30129
0.0944
Epoch  9, CIFAR-10 Batch 5:  2.3017
0.0944
Epoch 10, CIFAR-10 Batch 1:  2.3029
0.0974
Epoch 10, CIFAR-10 Batch 2:  2.30298
0.1008
Epoch 10, CIFAR-10 Batch 3:  2.30078
0.1002
Epoch 10, CIFAR-10 Batch 4:  2.30134
0.0944
Epoch 10, CIFAR-10 Batch 5:  2.30183
0.0944
Epoch 11, CIFAR-10 Batch 1:  2.30281
0.0978
Epoch 11, CIFAR-10 Batch 2:  2.30285
0.0976
Epoch 11, CIFAR-10 Batch 3:  2.3007
0.1002
Epoch 11, CIFAR-10 Batch 4:  2.30123
0.0944
Epoch 11, CIFAR-10 Batch 5:  2.30185
0.0944
Epoch 12, CIFAR-10 Batch 1:  2.30284
0.0978
Epoch 12, CIFAR-10 Batch 2:  2.30287
0.0976
Epoch 12, CIFAR-10 Batch 3:  2.30076
0.1004
Epoch 12, CIFAR-10 Batch 4:  2.30126
0.0946
Epoch 12, CIFAR-10 Batch 5:  2.30184
0.0946
Epoch 13, CIFAR-10 Batch 1:  2.30284
0.0978
Epoch 13, CIFAR-10 Batch 2:  2.30298
0.0974
Epoch 13, CIFAR-10 Batch 3:  2.30079
0.1002
Epoch 13, CIFAR-10 Batch 4:  2.30123
0.0942
Epoch 13, CIFAR-10 Batch 5:  2.30183
0.0944
Epoch 14, CIFAR-10 Batch 1:  2.30287
0.098
Epoch 14, CIFAR-10 Batch 2:  2.30292
0.0972
Epoch 14, CIFAR-10 Batch 3:  2.3007
0.1
Epoch 14, CIFAR-10 Batch 4:  2.30127
0.0942
Epoch 14, CIFAR-10 Batch 5:  2.30184
0.0942
Epoch 15, CIFAR-10 Batch 1:  2.30282
0.0978
Epoch 15, CIFAR-10 Batch 2:  2.30289
0.097
Epoch 15, CIFAR-10 Batch 3:  2.3007
0.0998
Epoch 15, CIFAR-10 Batch 4:  2.30122
0.0944
Epoch 15, CIFAR-10 Batch 5:  2.30186
0.0944
Epoch 16, CIFAR-10 Batch 1:  2.30285
0.0978
Epoch 16, CIFAR-10 Batch 2:  2.30295
0.097
Epoch 16, CIFAR-10 Batch 3:  2.30074
0.0998
Epoch 16, CIFAR-10 Batch 4:  2.30119
0.0942
Epoch 16, CIFAR-10 Batch 5:  2.30185
0.0942
Epoch 17, CIFAR-10 Batch 1:  2.30284
0.0978
Epoch 17, CIFAR-10 Batch 2:  2.30293
0.097
Epoch 17, CIFAR-10 Batch 3:  2.30071
0.0998
Epoch 17, CIFAR-10 Batch 4:  2.30123
0.0942
Epoch 17, CIFAR-10 Batch 5:  2.30185
0.0942
Epoch 18, CIFAR-10 Batch 1:  2.30284
0.0978
Epoch 18, CIFAR-10 Batch 2:  2.3029
0.097
Epoch 18, CIFAR-10 Batch 3:  2.3007
0.0998
Epoch 18, CIFAR-10 Batch 4:  2.30121
0.0942
Epoch 18, CIFAR-10 Batch 5:  2.30186
0.0942
Epoch 19, CIFAR-10 Batch 1:  2.30285
0.0978
Epoch 19, CIFAR-10 Batch 2:  2.3029
0.097
Epoch 19, CIFAR-10 Batch 3:  2.3007
0.0998
Epoch 19, CIFAR-10 Batch 4:  2.30121
0.0942
Epoch 19, CIFAR-10 Batch 5:  2.30185
0.0942
Epoch 20, CIFAR-10 Batch 1:  2.30285
0.0978
Epoch 20, CIFAR-10 Batch 2:  2.30291
0.097
Epoch 20, CIFAR-10 Batch 3:  2.3007
0.0998
Epoch 20, CIFAR-10 Batch 4:  2.30121
0.0942
Epoch 20, CIFAR-10 Batch 5:  2.30185
0.0942
Epoch 21, CIFAR-10 Batch 1:  2.30285
0.0978
Epoch 21, CIFAR-10 Batch 2:  2.30291
0.097
Epoch 21, CIFAR-10 Batch 3:  2.30071
0.0998
Epoch 21, CIFAR-10 Batch 4:  2.30121
0.0942
Epoch 21, CIFAR-10 Batch 5:  2.30185
0.0942
Epoch 22, CIFAR-10 Batch 1:  2.30285
0.0978
Epoch 22, CIFAR-10 Batch 2:  2.30291
0.097
Epoch 22, CIFAR-10 Batch 3:  2.30071
0.0998
Epoch 22, CIFAR-10 Batch 4:  2.30121
0.0942
Epoch 22, CIFAR-10 Batch 5:  2.30185
0.0942
Epoch 23, CIFAR-10 Batch 1:  2.30285
0.0978
Epoch 23, CIFAR-10 Batch 2:  2.30291
0.097
Epoch 23, CIFAR-10 Batch 3:  2.30071
0.0998
Epoch 23, CIFAR-10 Batch 4:  2.30121
0.0942
Epoch 23, CIFAR-10 Batch 5:  2.30185
0.0942
Epoch 24, CIFAR-10 Batch 1:  2.30285
0.0978
Epoch 24, CIFAR-10 Batch 2:  2.30291
0.097
Epoch 24, CIFAR-10 Batch 3:  2.30071
0.0998
Epoch 24, CIFAR-10 Batch 4:  2.30121
0.0942
Epoch 24, CIFAR-10 Batch 5:  2.30185
0.0942
Epoch 25, CIFAR-10 Batch 1:  2.30285
0.0978
Epoch 25, CIFAR-10 Batch 2:  2.30291
0.097
Epoch 25, CIFAR-10 Batch 3:  2.30071
0.0998
Epoch 25, CIFAR-10 Batch 4:  2.30121
0.0942
Epoch 25, CIFAR-10 Batch 5:  2.30185
0.0942
Epoch 26, CIFAR-10 Batch 1:  2.30285
0.0978
Epoch 26, CIFAR-10 Batch 2:  2.30291
0.097
Epoch 26, CIFAR-10 Batch 3:  2.30071
0.0998
Epoch 26, CIFAR-10 Batch 4:  2.30121
0.0942
Epoch 26, CIFAR-10 Batch 5:  2.30185
0.0942
Epoch 27, CIFAR-10 Batch 1:  2.30285
0.0978
Epoch 27, CIFAR-10 Batch 2:  2.30291
0.097
Epoch 27, CIFAR-10 Batch 3:  2.30071
0.0998
Epoch 27, CIFAR-10 Batch 4:  2.30121
0.0942
Epoch 27, CIFAR-10 Batch 5:  2.30185
0.0942
Epoch 28, CIFAR-10 Batch 1:  2.30285
0.0978
Epoch 28, CIFAR-10 Batch 2:  2.30291
0.097
Epoch 28, CIFAR-10 Batch 3:  2.30071
0.0998
Epoch 28, CIFAR-10 Batch 4:  2.30121
0.0942
Epoch 28, CIFAR-10 Batch 5:  2.30185
0.0942
Epoch 29, CIFAR-10 Batch 1:  2.30285
0.0978
Epoch 29, CIFAR-10 Batch 2:  2.30291
0.097
Epoch 29, CIFAR-10 Batch 3:  2.30071
0.0998
Epoch 29, CIFAR-10 Batch 4:  2.30121
0.0942
Epoch 29, CIFAR-10 Batch 5:  2.30185
0.0942
Epoch 30, CIFAR-10 Batch 1:  2.30285
0.0978
Epoch 30, CIFAR-10 Batch 2:  2.30291
0.097
Epoch 30, CIFAR-10 Batch 3:  2.30071
0.0998
Epoch 30, CIFAR-10 Batch 4:  2.30121
0.0942
Epoch 30, CIFAR-10 Batch 5:  2.30185
0.0942

Checkpoint

The model has been saved to disk.

Test Model

Test your model against the test dataset. This will be your final accuracy. You should have an accuracy greater than 50%. If you don't, keep tweaking the model architecture and parameters.


In [15]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
%config InlineBackend.figure_format = 'retina'

import tensorflow as tf
import pickle
import helper
import random

# Set batch size if not already set
try:
    if batch_size:
        pass
except NameError:
    batch_size = 64

save_model_path = './image_classification'
n_samples = 4
top_n_predictions = 3

def test_model():
    """
    Test the saved model against the test dataset
    """

    test_features, test_labels = pickle.load(open('preprocess_training.p', mode='rb'))
    loaded_graph = tf.Graph()

    with tf.Session(graph=loaded_graph) as sess:
        # Load model
        loader = tf.train.import_meta_graph(save_model_path + '.meta')
        loader.restore(sess, save_model_path)

        # Get Tensors from loaded model
        loaded_x = loaded_graph.get_tensor_by_name('x:0')
        loaded_y = loaded_graph.get_tensor_by_name('y:0')
        loaded_keep_prob = loaded_graph.get_tensor_by_name('keep_prob:0')
        loaded_logits = loaded_graph.get_tensor_by_name('logits:0')
        loaded_acc = loaded_graph.get_tensor_by_name('accuracy:0')
        
        # Get accuracy in batches for memory limitations
        test_batch_acc_total = 0
        test_batch_count = 0
        
        for train_feature_batch, train_label_batch in helper.batch_features_labels(test_features, test_labels, batch_size):
            test_batch_acc_total += sess.run(
                loaded_acc,
                feed_dict={loaded_x: train_feature_batch, loaded_y: train_label_batch, loaded_keep_prob: 1.0})
            test_batch_count += 1

        print('Testing Accuracy: {}\n'.format(test_batch_acc_total/test_batch_count))

        # Print Random Samples
        random_test_features, random_test_labels = tuple(zip(*random.sample(list(zip(test_features, test_labels)), n_samples)))
        random_test_predictions = sess.run(
            tf.nn.top_k(tf.nn.softmax(loaded_logits), top_n_predictions),
            feed_dict={loaded_x: random_test_features, loaded_y: random_test_labels, loaded_keep_prob: 1.0})
        helper.display_image_predictions(random_test_features, random_test_labels, random_test_predictions)


test_model()


Testing Accuracy: 0.10041799363057324

Why 50-70% Accuracy?

You might be wondering why you can't get an accuracy any higher. First things first, 50% isn't bad for a simple CNN. Pure guessing would get you 10% accuracy. However, you might notice people are getting scores well above 70%. That's because we haven't taught you all there is to know about neural networks. We still need to cover a few more techniques.

Submitting This Project

When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_image_classification.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.