vgg architecture

vgg_cifar.py

def convolutional_neural_network(x):
  weights = {'W_conv1' : tf.Variable(tf.random_normal([3,3,3,64])),
             'W_conv2' : tf.Variable(tf.random_normal([3,3,64,64])),
             'W_conv3' : tf.Variable(tf.random_normal([3,3,64,128])),
             'W_conv4' : tf.Variable(tf.random_normal([3,3,128,128])),
             'W_conv5': tf.Variable(tf.random_normal([3,3,128,256])),
             'W_conv6' : tf.Variable(tf.random_normal([3,3,256,256])),
             'W_conv7' : tf.Variable(tf.random_normal([3,3,256,256])),
             'W_conv8' : tf.Variable(tf.random_normal([3,3,256,512])),
             'W_conv9' : tf.Variable(tf.random_normal([3,3,512,512])),
             'W_conv10' : tf.Variable(tf.random_normal([3,3,512,512])),
             'W_fc1' : tf.Variable(tf.random_normal([2*2*512, 1024])),
             'W_fc2' : tf.Variable(tf.random_normal([1024, 1024])),
             'W_fc3' : tf.Variable(tf.random_normal([1024, n_classes]))}
  
  biases = {'b_conv1' : tf.Variable(tf.random_normal([64])),
            'b_conv2' : tf.Variable(tf.random_normal([64])),
            'b_conv3' : tf.Variable(tf.random_normal([128])),
            'b_conv4' : tf.Variable(tf.random_normal([128])),
            'b_conv5' : tf.Variable(tf.random_normal([256])),
            'b_conv6' : tf.Variable(tf.random_normal([256])),
            'b_conv7' : tf.Variable(tf.random_normal([256])),
            'b_conv8' : tf.Variable(tf.random_normal([512])),
            'b_conv9' : tf.Variable(tf.random_normal([512])),
            'b_conv10' : tf.Variable(tf.random_normal([512])),
            'b_fc1' : tf.Variable(tf.random_normal([1024])),
            'b_fc2' : tf.Variable(tf.random_normal([1024])),
            'b_fc3' : tf.Variable(tf.random_normal([n_classes]))}
  
  x = tf.reshape(x, [-1, 32, 32, 3])
  
  # Conv1
  conv1 = tf.nn.relu(conv2d(x, weights['W_conv1']) + biases['b_conv1'])
  
  # Conv2
  conv2 = tf.nn.relu(conv2d(conv1, weights['W_conv2']) + biases['b_conv2'])
  
  # Max pool
  conv2 = maxpool2d(conv2)
  
  # Conv3
  conv3 = tf.nn.relu(conv2d(conv2, weights['W_conv3']) + biases['b_conv3'])
  
  # Conv4
  
  conv4 = tf.nn.relu(conv2d(conv3, weights['W_conv4']) + biases['b_conv4'])
  
  # maxpool
  
  conv4 = maxpool2d(conv4)
  
  # Conv5
  conv5 = tf.nn.relu(conv2d(conv4, weights['W_conv5']) + biases['b_conv5'])
  
  #Conv6
  conv6 = tf.nn.relu(conv2d(conv5, weights['W_conv6']) + biases['b_conv6'])
  
  #Conv 7
  conv7 = tf.nn.relu(conv2d(conv6, weights['W_conv7']) + biases['b_conv7'])
  
  # max pool 
  conv7 = maxpool2d(conv7)
  
  #Conv 8
  conv8 = tf.nn.relu(conv2d(conv7, weights['W_conv8']) + biases['b_conv8'])
  
  #Conv9
  
  conv9 = tf.nn.relu(conv2d(conv8, weights['W_conv9']) + biases['b_conv9'])
  
  #Conv10
  conv10 = tf.nn.relu(conv2d(conv9, weights['W_conv10']) + biases['b_conv10'])
  
  # max pool
  conv10 = maxpool2d(conv10)
  
  #fc1 
  fc1 = tf.reshape(conv10, [-1, 2*2*512])
  fc1 = tf.nn.relu(tf.matmul(fc1, weights['W_fc1']) + biases['b_fc1'])
  fc1 = tf.nn.dropout(fc1, keep_rate)
  
  #fc2 
  fc2 = tf.nn.relu(tf.matmul(fc1, weights['W_fc2']) + biases['b_fc2'])
  fc2 = tf.nn.dropout(fc2, keep_rate)
  
  #Output Layer
  output = tf.nn.softmax(tf.matmul(fc2, weights['W_fc3']) + biases['b_fc3'])
  
  return output