Multi-module Spatial Semantic Network for Semantic Segmentation

In this work, we present an efficient method for semantic segmentation. In our method, a new network is introduced which consists of multiple modules to extract various features. Firstly, multi-scale spatial information is preserved in a spatial module with small stride and a pyramid pooling unit. Secondly, different level feature maps of backbone network are combined to extract multiple level semantic features. Finally, semantic features are further enhanced in the context module, in which various scales contextual information is integrated to facilitate complex scene understanding. We show competitive performance on two segmentation databases, including the CamVid and the Cityscapes dataset. The experimental results demonstrate that the proposed segmentation method can yield state-of-the-art segmentation accuracy. Contribution-We develop a multi-module network to implement semantic segmentation.