Refinement Co?supervision network for real?time semantic segmentation

Semantic segmentation is a fundamental technology for autonomous driving. It has high demand inference speed and accuracy. However, good trade-off between accuracy latency yet not present in existing semantic approaches. Due to the limitation of speed, authors cannot increase number network layers without limit design modules like networks real-time. challenging problem how model with performance under limited resources. To alleviate these issues, this study, propose refinement co-supervision (RCNet), which real-time on high-resolution image (1024×2048). The first construct context module, can provide low computation cost way obtaining large receptive field information. Furthermore, boundary mechanism proposed. strengthens optimisation easily neglected boundaries small targets. Experimental results reveal that proposed RCNet outperforms seven representative methods.