A Deep CNN based Kaggle Contest Winning Model to Recognize Real-Time Facial Expression

In face-to-face interactions, facial expressions convey nonverbal details. Since the early 1990s, researchers have been increasingly interested in automatic facial expression recognition, which is critical in human-machine interfaces. Traditional machine learning methods often necessitate a time-consuming procedure for identifying features and yield mediocre results. We propose efficient deep CNNs that are capable of accurately interpreting the semantic information available in faces without the need for hand-designing feature descriptors in an automated manner in this research paper, which makes use of recent advances in deep learning. We employ a range of loss functions and training techniques in order to learn CNNs with high classification efficiency. Using the renowned Facial Expressions Recognition (FER-2013) collection of data from the Kaggle facial expression recognition contest, the experimental findings show that our suggested networks outperform state-of-the-art techniques. On the contrary, to the competition’s winning model, our proposed networks’ parameter count is drastically reduced, resulting in an increase in average output speed of a significant magnitude and making the suggested networks a better fit for real-time applications.