Abstract
Real-time semantic segmentation plays an important role in practicalapplications such as self-driving and robots. Most research working on semanticsegmentation focuses on accuracy with little consideration for efficiency.Several existing studies that emphasize high-speed inference often cannotproduce high-accuracy segmentation results. In this paper, we propose a novelconvolutional network named Efficient Dense modules with Asymmetric convolution(EDANet), which employs an asymmetric convolution structure incorporating thedilated convolution and the dense connectivity to attain high efficiency at lowcomputational cost, inference time, and model size. Compared to FCN, EDANet is11 times faster and has 196 times fewer parameters, while it achieves a higherthe mean of intersection-over-union (mIoU) score without any additional decoderstructure, context module, post-processing scheme, and pretrained model. Weevaluate EDANet on Cityscapes and CamVid datasets to evaluate its performanceand compare it with the other state-of-art systems. Our network can run onresolution 512x1024 inputs at the speed of 108 and 81 frames per second on asingle GTX 1080Ti and Titan X, respectively.