We propose a novel encoder/decoder-based neural network architecture that learns view-dependent shape and appearance of geometry represented by voxel representations. Since the network is trained on local geometry patches, ...