model.py 文件源码

def model(inputs, is_training=False):
    # Multiple parallel convolutions
    block1 = multi_conv(inputs, [1, 2, 3, 4, 5, 6, 7, 8],
                        [512, 512, 512, 512, 512, 512, 512, 512],
                        name='block1', is_training=is_training)
    net = tf.reduce_max(block1, axis=1, name='maxpool')

    # Fully connected hidden layer (dense -> batch norm -> relu -> dropout)
    net = tf.layers.dense(net, 4096, kernel_regularizer=kernel_regularizer,
                          name='fc1')
    net = tf.layers.batch_normalization(net, training=is_training,
                                        name='fc1/batch_normalization')
    net = tf.nn.relu(net, name='fc1/relu')
    net = tf.layers.dropout(net, rate=0.5, training=is_training,
                            name='fc1/dropout')

    # Fully connected output layer
    net = tf.layers.dense(net, 4716, kernel_regularizer=kernel_regularizer,
                          name='fc2')
    tf.summary.histogram('summary/fc2', tf.nn.sigmoid(net))
    return net