Information Reuse Attention in Convolutional Neural Networks for Facial Expression Recognition in the Wild

Unlike the constraint frontal face condition, faces in the wild have various unconstrained interference factors, such as pose variations, illumination variations and occlusion. Because of this, facial expressions recognition (FER) in the wild is a challenging task and existing methods fail to performant well. However, for occluded faces (containing occlusion caused by other objects and self-occlusion caused by head posture changes), the attention mechanism has the ability to focus on the non-occluded regions automatically. In this paper, we propose an Information Reuse Attention Module (IRAM) for Convolutional Neural Network (CNN) to extract attention-aware features from faces. Our module reduces decay information in the process of generating attention maps by reusing the information of the previous layer and not reducing the dimensionality. Sequentially, we adaptively refine the feature responses by fusing the attention maps with the feature map. The proposed method is evaluated with two in-the-wild facial expression datasets RAF-DB and FER2013 and also compared with other state-of-the-art methods.