Haoliang Sheng, MengCheng Lau

Abstract: This study explores the performance of various ResNet architectures, namely ResNet18, ResNet34, and ResNet50, in the task of recognizing facial expressions using the FER-2013 dataset. By implementing transfer learning and fine-tuning these models, we rigorously evaluate their effectiveness through both controlled tests and real-time applications using webcam inputs. Our findings reveal that ResNet18, despite its simpler structure, achieves the optimal balance between accuracy and efficiency. This underscores its potential for practical deployment in real-time scenarios. The research emphasizes the critical importance of comprehensive datasets in enhancing model generalization, especially for expressions that are underrepresented in the training data. Additionally, it highlights the necessity of striking a careful balance between model complexity and computational efficiency to ensure suitability for real-world applications. This work significantly advances the field of real-time facial expression recognition, offering valuable insights into the deployment of deep learning models in practical, dynamic environments.

Keywords: Facial Expression Recognition, ResNet Architectures, Transfer Learning, Real-time Webcam.

Date Published: August 23, 2024
DOI: 10.11159/jmids.2024.005

View Article