In this paper, we propose an efficient architecture for semantic image segmentation using the depth-to-space (D2S) operation. Our D2S model is comprised of a standard CNN encoder followed by a depth-to-space reordering of the final convolutional feature maps; thus eliminating the decoder portion of traditional encoder-decoder segmentation models and reducing computation time almost by half. As a participant of the DeepGlobe Road Extraction competition, we evaluate our models on the corresponding road segmentation dataset. Our highly efficient D2S models exhibit comparable performance to standard segmentation models with much less computational cost.

翻译：在本文中,我们提出一个使用深度到空间(D2S)操作的语义图像分割高效结构。我们的D2S模型由标准的CNN编码器组成,然后对最后的进化地貌图进行深度到空间的重新排序,从而消除了传统编码器-解码分离模型的解码器部分,并将计算时间几乎减少一半。作为DeepGlobe路提取竞赛的参与者,我们评估了相应的路段分割数据集模型。我们高效的D2S模型的性能与标准分解模型相比,计算成本要低得多。