[Paper Review] Facial Emotion Recognition: State of the Art Performance on FER2013

Yousif Khaireddin, Zhuofa Chen|arXiv (Cornell University)|May 8, 2021

Emotion and Mood Recognition37 references158 citations

TL;DR

The paper achieves the highest single-network accuracy on FER2013 (73.28%) using a fine-tuned VGGNet without extra training data.

ABSTRACT

Facial emotion recognition (FER) is significant for human-computer interaction such as clinical practice and behavioral description. Accurate and robust FER by computer models remains challenging due to the heterogeneity of human faces and variations in images such as different facial pose and lighting. Among all techniques for FER, deep learning models, especially Convolutional Neural Networks (CNNs) have shown great potential due to their powerful automatic feature extraction and computational efficiency. In this work, we achieve the highest single-network classification accuracy on the FER2013 dataset. We adopt the VGGNet architecture, rigorously fine-tune its hyperparameters, and experiment with various optimization methods. To our best knowledge, our model achieves state-of-the-art single-network accuracy of 73.28 % on FER2013 without using extra training data.

Motivation & Objective

• Motivate improved FER performance due to the heterogeneity of faces and image conditions.
• Demonstrate that a carefully tuned VGGNet can achieve state-of-the-art accuracy on FER2013 without external data.
• Investigate the impact of hyperparameter fine-tuning and optimization strategies on FER performance.

Proposed method

• Adopted the VGGNet architecture as the backbone.
• Rigorous hyperparameter fine-tuning of the network.
• Experimented with various optimization methods.
• Focused on achieving high performance with a single network.
• Emphasized no use of extra training data beyond FER2013.

Experimental results

Research questions

• RQ1 Can a single CNN network achieve state-of-the-art FER performance on FER2013 without external data?
• RQ2 What is the impact of hyperparameter tuning and optimization methods on FER2013 accuracy?
• RQ3 Is VGGNet a viable backbone for top FER performance with careful training?

Key findings

• Achieved 73.28% accuracy on FER2013 with a single network.
• Reported as the highest single-network FER2013 result without additional data.
• Validated that hyperparameter tuning and optimization choices can yield state-of-the-art performance.
• Demonstrated effectiveness of the VGGNet backbone for FER when properly tuned.