Light-Weight Semantic Segmentation Based on Mask RCNN

Abstract

Semantic segmentation is based on the image information to detect and identify various categories of objects and output these objects mask. Compared with object detection and image classification, semantic segmentation has a more accurate recognition effect, and has a wide range of applications in automatic driving and other fields. With the development of deep learning, semantic segmentation methods based on deep learning have achieved good results, such as mask rcnn. Nowadays, Mask RCNN already has a good ability to handle object segmentation and mask the object. However, the model needs to consume a lot of computational power so that raise request for the equipment. This paper proposes a lightweight semantic segmentation network based on Mask RCNN, which can achieve better semantic segmentation accuracy with less data. In particular, using deep convolution as the feature extraction operator can effectively reduce the calculation and parameter number of the model. Experiments show the effectiveness of the proposed method.