Facial expression recognition in the wild: A new Adaptive Attention-Modulated Contextual Spatial Information network

Abstract

Facial expression recognition (FER) is an important and widely applied task. This paper proposes an Adaptive Attention-modulated Contextual Spatial Information (Ad-ACSI) model to improve FER in uncontrolled environments. The proposed Ad-ACSI model incorporates an Attention-modulated Contextual Spatial Information Network (ACSI-Net), a joint loss, and an adaptive attention modulator (AAM). The ACSI-Net, built on ResNet with contextual convolution (CoConv) and coordinated attention (CA), effectively captures global and local contextual features. The adaptive attention modulator (AAM) refines the features and generates dynamic weights for the center loss. The cross-entropy (CE) loss is refined into an equilibrium loss and combined with a sparse center loss to improve inter-class discrimination and intra-class clustering. Experiments on the RAF-DB and AffectNet datasets show that the proposed method achieves results comparable to state-of-the-art methods of FER in the wild, with it promising for integration into popular architectures.