Facial variation-aware Learning Transformer for Dynamic Facial Expression Recognition

Dynamic Facial Expression Recognition (DFER) is an indispensable upstream task in the domains of human-machine interaction, driving safety, education, and healthcare aids. However, the inherent uncertainty between variations significantly hinders the effectiveness of DFER in practical applications, primarily due to the similar facial changes associated with different facial expressions. To address this challenge, we introduced a novel relationship-driven approach based on the Transformer architecture, which can learn the relationships between facial expressions and facial variations. Specifically, we designed several facial variation tokens to explicitly encode the variation status of key facial areas. Additionally, we designed a novel module to model the interactions between facial variation tokens, helping the model better focus on the specific facial variations of expressions. Experimental results on two challenging benchmark DFER datasets demonstrate that our proposed FVATrans achieves state-of-the-art performance.