def inference(images, keep_prob):

  # 416 x 416 x 3
  with tf.variable_scope('downSample_block1') as scope:
    conv1 = Conv2D(filters=64, kernel_size=(3, 3), strides=(1, 1),
     padding='valid', name='conv1', activation='relu')(images)
    conv2 = Conv2D(filters=64, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv2', activation='relu')(conv1)
    pool1 = MaxPool2D(pool_size=(2, 2), strides=(2, 2), padding='valid', name='pool')(conv2)

  with tf.variable_scope('downSample_block2') as scope:
    conv3 = Conv2D(filters=128, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv1', activation='relu')(pool1)
    conv4 = Conv2D(filters=128, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv2', activation='relu')(conv3)
    pool2 = MaxPool2D(pool_size=(2, 2), strides=(2, 2), padding='valid', name='pool')(conv4)

  with tf.variable_scope('downSample_block3') as scope:
    conv5 = Conv2D(filters=256, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv1', activation='relu')(pool2)
    conv6 = Conv2D(filters=256, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv2', activation='relu')(conv5)
    pool3 = MaxPool2D(pool_size=(2, 2), strides=(2, 2), padding='valid', name='pool')(conv6)

  with tf.variable_scope('downSample_block4') as scope:
    conv7 = Conv2D(filters=512, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv1', activation='relu')(pool3)
    conv8 = Conv2D(filters=512, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv2', activation='relu')(conv7)
    pool4 = MaxPool2D(pool_size=(2, 2), strides=(2, 2), padding='valid', name='pool')(conv8)

  with tf.variable_scope('downSample_block5') as scope:
    conv9 = Conv2D(filters=1024, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv1', activation='relu')(pool4)
    conv10 = Conv2D(filters=1024, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv2', activation='relu')(conv9)
    drop1 = tf.nn.dropout(conv10, keep_prob)

  with tf.variable_scope('upSample_block1') as scope:
    deconv1 = Conv2DTranspose(filters=512, kernel_size=(2, 2), strides=(2, 2),
      padding='same', name='deconv')(drop1)
    crop1 = 4
    crop2 = 40
    concat_deconv1 = tf.concat([conv8[:, crop1 : crop2 , crop1 : crop2, :], deconv1],axis=3)
    conv11 = Conv2D(filters=512, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv1', activation='relu')(concat_deconv1)
    conv12 = Conv2D(filters=512, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv2', activation='relu')(conv11)

  with tf.variable_scope('upSample_block2') as scope:
    deconv2 = Conv2DTranspose(filters=256, kernel_size=(2, 2), strides=(2, 2),
      padding='same', name='deconv')(conv12)
    crop1 = 16
    crop2 = 80
    concat_deconv2 = tf.concat([conv6[:, crop1 : crop2 , crop1 : crop2, :], deconv2],axis=3)
    conv13 = Conv2D(filters=256, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv1', activation='relu')(concat_deconv2)
    conv14 = Conv2D(filters=256, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv2', activation='relu')(conv13)

  with tf.variable_scope('upSample_block3') as scope:
    deconv3 = Conv2DTranspose(filters=128, kernel_size=(2, 2), strides=(2, 2),
      padding='same', name='deconv')(conv14)
    crop1 = 40
    crop2 = 160
    concat_deconv3 = tf.concat([conv4[:, crop1 : crop2 , crop1 : crop2, :], deconv3],axis=3)
    conv15 = Conv2D(filters=128, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv1', activation='relu')(concat_deconv3)
    conv16 = Conv2D(filters=128, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv2', activation='relu')(conv15)

  with tf.variable_scope('upSample_block4') as scope:
    deconv4 = Conv2DTranspose(filters=64, kernel_size=(2, 2), strides=(2, 2),
      padding='same', name='deconv')(conv16)
    crop1 = 89
    crop2 = 321
    concat_deconv4 = tf.concat([conv2[:, crop1 : crop2 , crop1 : crop2, :], deconv4],axis=3)
    conv17 = Conv2D(filters=64, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv1', activation='relu')(concat_deconv4)
    conv18 = Conv2D(filters=64, kernel_size=(3, 3), strides=(1, 1),
      padding='valid', name='conv2', activation='relu')(conv17)
    drop2 = tf.nn.dropout(conv18, keep_prob)

  with tf.variable_scope('Classification') as scope:
    conv19 = Conv2D(filters=2, kernel_size=(1, 1), strides=(1, 1),
      padding='valid', name='conv1', activation='relu')(drop2)

  with tf.variable_scope('softmax') as scope:
    softmax = soft_max(conv19)

  # Output 228 x 228 x 2

  return softmax