Multiple Resolutions Detail Enhancement Network for Real-Time Image Semantic Segmentation

Real-time image semantic segmentation (ISS) draws the attentions of more and more researchers as a basis of scene understanding, and it has been applied in many fields that need fast interaction and response, such as autonomous driving and robot control. Considering the loss of low-level spatial information with the deepening network layer, we propose a multiple resolutions detail enhancement network (MRDENet) in this article, which adequately extracts and utilizes accurate low-level detail information from original images with different resolutions. MRDENet consists of three light-weight branch subnetworks, and designs dense oblique connections between adjacent branches to preserve the different level effective features of previous branch. Furthermore, a new multilevel information aggregation (MIA) module is presented to effectively fuse the low-level detail features and the high-level semantic features of different branches by employing group convolution (group conv) and channel shuffle with low computation cost, thus ensuring that MRDENet could achieve a favorable tradeoff between segmentation precision with inference speed. The experimental results show that MRDENet achieves 73.1% mean intersection over union (mIoU) with 93 frames per second (FPS) on Cityscapes dataset, and 68.5% mIoU with 112 FPS on CamVid dataset, which indicates the performance of MRDENet is competitive with the state-of-art methods.