Concentric Ring Loss for Face Forgery Detection

Yu Yin, Yue Bai, Yun Fu

Keywords: face forgery detection, metric learning

Abstract: Due to growing societal concerns about indistinguishable deepfake images, face forgery detection has received an increasing amount of interest in computer vision. Since the differences between actual and fake images are frequently small, improving the discriminative ability of learnt features is one of the primary problems in deepfake detection. In this paper, we propose a novel Concentric Ring Loss (CRL) to encourage the model to learn intra-class compressed and inter-class separated features. Specifically, we independently add margin penalties in angular and Euclidean space to force a more significant margin between real and fake images, and hence encourage better discriminating performance. Compared to softmax loss, CRL explicitly encourages intra-class compactness and inter-class separability. Moreover, a frequency-aware feature learning module is proposed to exploit high-frequency features and further improve the generalization ability of the model. Extensive experiments demonstrate the superiority of our methods over different datasets. We show that CRL consistently outperforms the state-of-the-art by a large margin.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Deep Learning and representational learning