main.py

#!/usr/bin/env python3
import os.path
import tensorflow as tf
import helper 
import warnings
from distutils.version import LooseVersion
import project_tests as tests


# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer.  You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))

# Check for a GPU
if not tf.test.gpu_device_name():
    warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))


def load_vgg(sess, vgg_path):
    """
    Load Pretrained VGG Model into TensorFlow.
    :param sess: TensorFlow Session
    :param vgg_path: Path to vgg folder, containing "variables/" and "saved_model.pb"
    :return: Tuple of Tensors from VGG model (image_input, keep_prob, layer3_out, layer4_out, layer7_out)
    """
    # TODO: Implement function
    #   Use tf.saved_model.loader.load to load the model and weights
    vgg_tag = 'vgg16'
    vgg_input_tensor_name = 'image_input:0'
    vgg_keep_prob_tensor_name = 'keep_prob:0'
    vgg_layer3_out_tensor_name = 'layer3_out:0'
    vgg_layer4_out_tensor_name = 'layer4_out:0'
    vgg_layer7_out_tensor_name = 'layer7_out:0'
    
    tf.saved_model.loader.load(sess,[vgg_tag],vgg_path)
    graph = tf.get_default_graph() # get the default graph
    w1 = graph.get_tensor_by_name(vgg_input_tensor_name)
    keep = graph.get_tensor_by_name(vgg_keep_prob_tensor_name)
    layer3 = graph.get_tensor_by_name(vgg_layer3_out_tensor_name)
    layer4 = graph.get_tensor_by_name(vgg_layer4_out_tensor_name)
    layer7 = graph.get_tensor_by_name(vgg_layer7_out_tensor_name)

    return w1,keep, layer3, layer4,layer7
tests.test_load_vgg(load_vgg, tf)


def layers(vgg_layer3_out, vgg_layer4_out, vgg_layer7_out, num_classes):
    """
    Create the layers for a fully convolutional network.  Build skip-layers using the vgg layers.
    :param vgg_layer3_out: TF Tensor for VGG Layer 3 output
    :param vgg_layer4_out: TF Tensor for VGG Layer 4 output
    :param vgg_layer7_out: TF Tensor for VGG Layer 7 output
    :param num_classes: Number of classes to classify
    :return: The Tensor for the last layer of output
    """
    # TODO: Implement function

    l7_conv_1x1 = tf.layers.conv2d(vgg_layer7_out, num_classes, 1, padding='same', kernel_initializer=tf.truncated_normal_initializer(stddev=0.01), kernel_regularizer=tf.contrib.layers.l2_regularizer(0.001))          # reduce the number of filters from 4096 down to num_classes
    l7_output = tf.layers.conv2d_transpose(l7_conv_1x1,num_classes, 4,2,padding='same', kernel_initializer=tf.truncated_normal_initializer(stddev=0.01), kernel_regularizer=tf.contrib.layers.l2_regularizer(0.001))     # increase the height and width dimensions of the input tensor

    l4_scaled = tf.multiply(vgg_layer4_out, 0.01)  # the l4 layer on the encoder is scaled so we have to scale it back
    l4_conv_1x1 = tf.layers.conv2d(l4_scaled, num_classes, 1, padding='same', kernel_initializer=tf.truncated_normal_initializer(stddev=0.01), kernel_regularizer=tf.contrib.layers.l2_regularizer(0.001))    
    l4_skip = tf.add(l7_output, l4_conv_1x1) # add skip layer
    l4_output = tf.layers.conv2d_transpose(l4_skip,num_classes, 4,2,padding='same', kernel_initializer=tf.truncated_normal_initializer(stddev=0.01), kernel_regularizer=tf.contrib.layers.l2_regularizer(0.001))
    
    l3_scaled = tf.multiply(vgg_layer3_out, 0.0001) # the l3 layer on the encoder is scaled so we have to scale it back
    l3_conv_1x1 = tf.layers.conv2d(l3_scaled, num_classes, 1, padding='same', kernel_initializer=tf.truncated_normal_initializer(stddev=0.01), kernel_regularizer=tf.contrib.layers.l2_regularizer(0.001))
    l3_skip = tf.add(l3_conv_1x1, l4_output) # add skip layer   
    output = tf.layers.conv2d_transpose(l3_skip,num_classes, 16,8 ,padding='same', kernel_initializer=tf.truncated_normal_initializer(stddev=0.01), kernel_regularizer=tf.contrib.layers.l2_regularizer(0.001))   

    return output
tests.test_layers(layers)


def optimize(nn_last_layer, correct_label, learning_rate, num_classes):
    """
    Build the TensorFLow loss and optimizer operations.
    :param nn_last_layer: TF Tensor of the last layer in the neural network
    :param correct_label: TF Placeholder for the correct label image
    :param learning_rate: TF Placeholder for the learning rate
    :param num_classes: Number of classes to classify
    :return: Tuple of (logits, train_op, cross_entropy_loss)
    """
    # TODO: Implement function
    logits = tf.reshape(nn_last_layer, (-1,num_classes))
        #####  correct_label = tf.reshape(correct_label, (-1,num_classes))
    cross_entropy_loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits = logits, labels = correct_label))     # define the loss function 
    regularization_loss = tf.get_collection(tf.GraphKeys.REGULARIZATION_LOSSES)     # get the regularization losses

    total_loss = cross_entropy_loss + sum(regularization_loss) # calculate the total loss
    adam_optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate)    # define the optimizer    
    train_op = adam_optimizer.minimize(cross_entropy_loss)                  # Define train operation
    return logits, train_op, total_loss
tests.test_optimize(optimize)


def train_nn(sess, epochs, batch_size, get_batches_fn, train_op, cross_entropy_loss, input_image,
             correct_label, keep_prob, learning_rate):
    """
    Train neural network and print out the loss during training.
    :param sess: TF Session
    :param epochs: Number of epochs
    :param batch_size: Batch size
    :param get_batches_fn: Function to get batches of training data.  Call using get_batches_fn(batch_size)
    :param train_op: TF Operation to train the neural network
    :param cross_entropy_loss: TF Tensor for the amount of loss
    :param input_image: TF Placeholder for input images
    :param correct_label: TF Placeholder for label images
    :param keep_prob: TF Placeholder for dropout keep probability
    :param learning_rate: TF Placeholder for learning rate
    """
    # TODO: Implement function
    print("\ntrain_nn started\n\n")
    sess.run(tf.global_variables_initializer())
    for epoch in range(epochs):
        print ("Epoch: ", epoch, "\n")
        for image, label in get_batches_fn(batch_size):
            _, loss = sess.run([train_op, cross_entropy_loss], feed_dict={
                        input_image: image,
                        correct_label: label,
                        keep_prob: 0.5,                        
                        learning_rate: 0.0001 })
            print("loss: ",loss,"\n")
            if loss <0.1:  # early exit if  the model is trained good engough, it saves time and it prevents overfitting
                print ("The loss is good enough, exit training.")
                return
tests.test_train_nn(train_nn)


def run():
    create_video = True  # requires imageio package. It creates a video from the processed images
    num_classes = 3    
    # I tried to implement a version with three classes but it did not work. 
    # The model could not make a distinction between the current road and other roads on the images. 
    # I used green color for the current road and blue color for the other roads. 
    # The other road was colored but there were mixed color pixels on the edges.   
    # I included the helper_3classes.py file. It includes the code with the modifications I made to test three classes detection.                                                                                                                    
    image_shape = (160, 576)  # KITTI dataset uses 160x576 images
    data_dir = './data'
    runs_dir = './runs'
    epochs = 30
    batch_size = 20 # It is tested on a GPU with 11GB RAM. Reduce this value if you get an error.
    tests.test_for_kitti_dataset(data_dir)

    # Download pretrained vgg model
    helper.maybe_download_pretrained_vgg(data_dir)

    # OPTIONAL: Train and Inference on the cityscapes dataset instead of the Kitti dataset.
    # You'll need a GPU with at least 10 teraFLOPS to train on.
    #  https://www.cityscapes-dataset.com/

    with tf.Session() as sess:
        # Path to vgg model
        vgg_path = os.path.join(data_dir, 'vgg')
        # Create function to get batches
        get_batches_fn = helper.gen_batch_function(os.path.join(data_dir, 'data_road/training'), image_shape)

        # OPTIONAL: Augment Images for better results
        #  https://datascience.stackexchange.com/questions/5224/how-to-prepare-augment-images-for-neural-network

        # TODO: Build NN using load_vgg, layers, and optimize function
        input_image, keep_prob, layer3_out, layer4_out, layer7_out = load_vgg(sess, vgg_path)   # load the saved vgg model and create the layers of the NN
        layer_output = layers(layer3_out,layer4_out, layer7_out, num_classes)                   # create the output layer

        correct_label = tf.placeholder(tf.int32, [None, None, None, num_classes])
        learning_rate = tf.placeholder(tf.float32)
        logits, train_op, cross_entropy_loss = optimize(layer_output,correct_label, learning_rate, num_classes)
        # TODO: Train NN using the train_nn function

        train_nn(sess, epochs, batch_size, get_batches_fn, train_op, cross_entropy_loss, input_image,correct_label, keep_prob, learning_rate)    # train the NN
        # TODO: Save inference data using helper.save_inference_samples
        helper.save_inference_samples(runs_dir, data_dir, sess, image_shape, logits, keep_prob, input_image, create_video )

        # OPTIONAL: Apply the trained model to a video


if __name__ == '__main__':
    run()