Architecture of Deep Convolutional Encoder-Decoder Networks for Building Footprint Semantic Segmentation