Semantic Binary Segmentation using Convolutional Networks without Decoders
This work addresses computational efficiency in segmentation for applications like road extraction, but it is incremental as it modifies existing encoder-decoder frameworks.
The paper tackles semantic image segmentation by proposing an efficient architecture that replaces the decoder with a depth-to-space operation, reducing computation by almost half while achieving comparable performance on a road segmentation dataset.
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. Our approach eliminates the decoder portion of traditional encoder-decoder segmentation models and reduces the amount of computation 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 lower computational cost.