Research on Clothing Image Segmentation based on EF-UNet
An improved VGG16-UNet garment image segmentation method(EF-UNet)is proposed to address the problems of low accuracy of localized garment segmentation,as well as the occurrence of mis-segmentation and o-mission.By adding the ECA attention mechanism at the end of the encoder,the encoder layer 5 is given more weight to better extract the target feature infor-mation,thus enhancing the segmentation ability of the garment im-age.Further,the addition of multi-level feature fusion is used in the decoder to capture features of various scales to achieve the purpose of improving the segmentation effect.The results show that the improved segmentation model has a higher model training index and better segmentation effect compared with UNet,PSPNet,DeepLab v3+,and VGG16-UNet semantic segmentation models.Compared with VGG16-UNet mean intersection over union(mIOU),mean pixel accuracy(mPA)and accuracy(Accuracy)are improved by 4.62％,4.59％and 0.29％,respective-ly.
万方数据
维普
