Locally GAN-generated face detection based on an improved Xception

Abstract It has become a research hotspot to detect whether a face is natural or GAN-generated. However, all the existing works focus on whole GAN-generated faces. So, an improved Xception model is proposed for locally GAN-generated face detection. To the best of our knowledge, our work is the first one to address this issue. Some improvements over Xception are as follows: (1) Four residual blocks are removed to avoid the overfitting problem as much as possible; (2) Inception block with the dilated convolution is used to replace the common convolution layer in the pre-processing module of the Xception to obtain multi-scale features; (3) Feature pyramid network is utilized to obtain multi-level features for final decision. The first locally GAN-based generated face (LGGF) dataset is constructed by the pluralistic image completion method on the basis of FFHQ dataset. It has a total 952,000 images with the generated regions in different shapes and sizes. Experimental results demonstrate the superiority of the proposed model which outperforms some existing models, especially for the faces having small generated regions.