Research on Semantic Segmentation of Portraits Based on Improved Deeplabv3 +. Issue 1 (April 2020)

• Record Type: Journal Article
• Authors: Zhang, Kaihui; Liu, Xianhui; Chen, Yufei
• Is Part Of: IOP conference series. Volume 806:Issue 1(2020)
• Publication Date: 2020-04
• Subjects: Materials science -- Periodicals; 620.1105
• DOI: 10.1088/1757-899X/806/1/012057
• Languages: English
• ISSNs: 1757-8981

Abstract

Abstract: At present, semantic segmentation is one of the hottest research directions in the field of computer vision, and the Deeplab series is one of the best neural network models in semantic segmentation. Based on the original Deeplab-v3 + neural network model, this paper makes a series of improvements to make it perform better on semantic segmentation of portraits: 1. Add channel attention module to speed up the convergence speed of model training and improve segmentation accuracy. 2. Improve the original AtrousSpatial Pyramid Pooling (ASPP), adjust the receptive field of the network, and improve the segmentation accuracy. Finally, this paper compares the segmentation performance of different semantic segmentation network models on human-parsing dataset and proves that the proposed neural network structure has better segmentation effect.