A Method for Human Parsing Based on Deep Learning and Attention Mechanism

Abstract

Human parsing is thought as a specific image semantic segmentation task. Most existing methods adopt encoder- decoder framework, and make full use of global context to achieve better image segmentation effect. A model is proposed in this paper which uses Convolutional Block Attention Module to select more discriminate feature and refinement residual learning to learn residual representation between input and output. The experiment results shows that our proposed model can achieve better performance on our dataset than other networks. From the generated human body segmentation images, the model can achieve more details and semantic consistency.