Usage of ResNet18 with CBAM Attention Mechanisms in Facial Emotion Recognition

Facial Emotion Recognition (FER) is a pivotal task in computer vision and is increasingly being powered by Artificial Intelligence (AI) techniques. These AI -driven applications span areas such as mental illness detection, biometrics, and psychological profiling. One of the most influential AI subsets employed in FER is deep learning, especially Convolutional Neural Networks (CNNs). The current state-of-the-art (SOTA) single-network classification accuracy on the FER-2013 dataset, achieved without additional training data, is attributed to the Visual Geometry Group (VGG) model, with a performance rate of 73.28%. In this work, we harness the capabilities of ResNet18 integrated with CBAM attention, achieving a test accuracy of 73.67% without the need for extra training data and also enhancing the convergence speed.