 This paper proposes a novel deep learning architecture for semantic segmentation of aerial images. The proposed model uses deconvolutional layers and recycled early layers to convert intensity and range data into a pixel level classification at full resolution. The authors show that an ensemble of multiple models can achieve excellent results on the ISPRS Semantic Labeling Benchmark dataset, demonstrating the effectiveness of the proposed approach. This article was authored by D. Marmani's, J. D. Wagner, S. Galliani, and others.