Mini-ResEmoteNet: Leveraging Knowledge Distillation for Human-Centered Design

Facial Emotion Recognition has emerged as increasingly pivotal in the domainof User Experience, notably within modern usability testing, as it facilitatesa deeper comprehension of user satisfaction and engagement. This study aims toextend the ResEmoteNet model by employing a knowledge distillation framework todevelop Mini-ResEmoteNet models - lightweight student models - tailored forusability testing. Experiments were conducted on the FER2013 and RAF-DBdatasets to assess the efficacy of three student model architectures: StudentModel A, Student Model B, and Student Model C. Their development involvesreducing the number of feature channels in each layer of the teacher model byapproximately 50%, 75%, and 87.5%. Demonstrating exceptional performance on theFER2013 dataset, Student Model A (E1) achieved a test accuracy of 76.33%,marking a 0.21% absolute improvement over EmoNeXt. Moreover, the resultsexhibit absolute improvements in terms of inference speed and memory usageduring inference compared to the ResEmoteNet model. The findings indicate thatthe proposed methods surpass other state-of-the-art approaches.