基于深度学习的半监督语义分割方法

his paper proposes a semi supervised semantic segmentation method G-Face to address the issues of inaccurate facial feature localization in current face detection algorithms, resulting in low detection model accuracy and high time cost due to manually annotated features. Based on this, a semi supervised semantic segmentation model based on generative adversarial networks is proposed. By combining current popular forgery detection methods and G-Face, experiments were conducted on the Celeb-DF public dataset and the constructed dataset in this paper to compare and analyze whether this method can improve the performance of the detection model. The experimental results show that the proposed method can effectively improve the accuracy of the detection model in recognizing facial images, and the performance evaluation indicators of the evaluation model are all improved by about 5%.