Research And Implementation Of Light Weighted Semantic Segmentation Based On Ege Information

Semantic segmentation is one of the most challenging tasks in the field of computer vision.With the rise of the hottest topics in the field of computer vision such as scene understanding,reconstruction,and image processing,image semantic segmentation,as the basis of these hot topics,has also received more and more attention from researchers in this field.However,most of the current semantic segmentation algorithms achieve better results at the cost of computing resources.This has led to a huge problem when the algorithm is implemented:excessive computation and parameter volume increase the overall cost of the algorithm.On the other hand,it also greatly increases the threshold for using the technology,which greatly limits the application scenarios of the algorithm.Thus,the need for lightweight semantic segmentation algorithms is imminent.In the view of above problems,this paper introduces edge information into the deep learning-based semantic segmentation method,so as to realize the lightweight of the semantic segmentation algorithm.The algorithm designs an efficient semantic segmentation framework,which includes three main parts:efficient edge detection branch,efficient semantic segmentation branch and multi-layer edge semantic information fusion module.Our work propose a more efficient edge detection branch and semantic segmentation branch based on existing lightweight backbone networks.The edge information is better introduced into the semantic information through the multi-layer edge semantic information fusion module,so that the algorithm is significantly improved in efficiency.This paper also designs and implements an image stylization system based on semantic segmentation.The system design is mainly divided into two phases and three modules.Our system mainly includes the offline training and online inference phases.The offline training phase includes a pre-training module and a joint training module.And the online reasoning phase includes a mobile-side inference module.To sum up,this paper analyzes the shortcomings of the existing methods of semantic segmentation.Based on these observations,we propose a lightweight semantic segmentation algorithm based on image edge information.We further design and implement an image stylization system based on above semantic segmentation algorithm.The comparison experiments and system tests on the public dataset Cityscapes verify the effectiveness of the proposed method and the stability and usability of the designed system.