Local Multi-Head Channel Self-Attention for Facial Expression Recognition

Since the Transformer architecture was introduced in 2017, there has been many attempts to bring self-attention paradigm field of computer vision. In this paper, we propose LHC: Local multi-Head Channel self-attention, a novel module that can be easily integrated into virtually every convolutional neural network, and is specifically designed for vision, with specific focus on facial expression recognition. LHC based two main ideas: first, think best way leverage channel-wise application instead more well explored spatial attention. Secondly, local approach potential better overcome limitations convolution than global attention, at least those scenarios where images have constant general structure, as LHC-Net achieves new state-of-the-art FER2013 dataset, significantly lower complexity impact “host” terms computational cost when compared previous state-of-the-art.