Multi-Attention Module for Dynamic Facial Emotion Recognition

Video-based dynamic facial emotion recognition (FER) is a challenging task, as one must capture and distinguish tiny movements representing emotional changes while ignoring the differences of different objects. Recent state-of-the-art studies have usually adopted more complex methods to solve this such large-scale deep learning models or multimodal analysis with reference multiple sub-models. According characteristics FER task shortcomings existing methods, in paper we propose lightweight method design three attention modules that can be flexibly inserted into backbone network. The key information for dimensions space, channel, time extracted by means convolution layer, pooling multi-layer perception (MLP), other approaches, weights are generated. By sharing parameters at same level, do not add too many network enhancing focus on specific areas face, effective feature static images, frames. experimental results CK+ eNTERFACE’05 datasets show achieve higher accuracy.