Enhanced Emotion Recognition on the FER-2013 Dataset by Training VGG from Scratch

Muhammad Talha Jahangir ()
Additional contact information
Muhammad Talha Jahangir: Department of Computer Science, MNS-University of Engineering and Technology, Multan, Pakistan

International Journal of Innovations in Science & Technology, 2024, vol. 6, issue 4, 1691-1719

Abstract: Recognizing facial emotions is still a major obstacle in computer vision, particularly when dealing with complex datasets such as FER-2013. Advancements in deep learning have simplified the process of achieving high accuracy, yet obtaining high accuracy on the FER-2013 dataset with traditional methods remains challenging. The aim of this research is to analyze the effectiveness of Convolutional Neural Networks (CNNs) utilizing VGG16 and ResNet50 architectures through three different training methods: training from scratch, with transfer learning, and fine-tuning. Our research demonstrates that while training VGG16 from scratch achieved a validation accuracy of 67.23%, fine-tuning produced a slight reduction in performance at 64.80%. Conversely, ResNet50 struggled across all approaches, with the highest validation accuracy being only 54.69% when trained from scratch. We offer an in-depth analysis of these methodologies by utilizing confusion matrices, training durations, and accuracy measures to showcase the balance between computational expenses and model effectiveness. Our results indicate that, although transfer learning and fine-tuning offer rapid convergence, training from scratch may still be necessary for specialized feature learning in complex FER tasks. These results help in the continuous work of improving emotion recognition systems by maintaining a balance between accuracy and computational efficiency.

Keywords: Facial Emotion Recognition; Convolutional; Neural Network; FER-2013 Dataset; Pre-Trained Models; Fine Tuning; VGG16; ResNet50 (search for similar items in EconPapers)
Date: 2024
References: Add references at CitEc
Citations:

Downloads: (external link)
https://journal.50sea.com/index.php/IJIST/article/view/1085/1617 (application/pdf)
https://journal.50sea.com/index.php/IJIST/article/view/1085 (text/html)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:abq:ijist1:v:6:y:2024:i:4:p:1691-1719

Access Statistics for this article

International Journal of Innovations in Science & Technology is currently edited by Prof. Dr. Syed Amer Mahmood

More articles in International Journal of Innovations in Science & Technology from 50sea
Bibliographic data for series maintained by Iqra Nazeer ().