Temporal Facial Alignment with Triple Discriminators

Seongmin Lee, Hyunse Yoon, Sanghoon Lee, Jiwoo Kang

Abstract: Facial alignment is one of the most important features of 3D facial models. However, when fitting the 3D Morphable Model (3DMM) into a large expression or pose facial image, there is ambiguity as to whether facial shape deformation is caused by identity or expression. To address this, in this paper, we propose a stable and accurate facial alignment framework by introducing multiple stability discriminators. The proposed discriminators determine the regressed camera, face identity, and expression parameters simultaneously from an image. The proposed framework for facial alignment consists of a facial alignment network and stability discriminators: identity, expression, and temporal discriminators. The facial alignment network is trained to predict camera, face identity, and expression parameters based on an image. The stability discriminator is trained to distinguish whether the facial deformation generated by the estimated facial identity and expression parameters is stable. Meanwhile, the discriminator distinguishes whether the deformation between adjacent frames is consistent. By utilizing these stability discriminators, the proposed facial alignment network demonstrates precise and consistent performance in aligning faces in both static and dynamic domains. To verify the performance of the proposed discriminators, the large-scale facial tracking dataset, 300 Videos in the Wild (300VW) dataset, is used for qualitative and quantitative evaluations. The experimental results show that the proposed method outperforms state-of-the-art methods, demonstrating the strong benefits of our method in accurate facial alignment over time.