Scholarlink

<p>Facial expressions are an important channel for interpersonal communication and comprehension since people externalize their emotions through a variety of facial expressions. Technology, in particular, deep learning algorithms, can detect and analyze human emotions in real-time, which paves the way for advanced user interfaces or adjustable devices and applications. Based on this, the work described in this paper presents a system that identifies three groups of emotions, positive, negative, and neutral, in three execution scenarios: in resource-limited devices, such as mobile phones, for desktop or web applications, and with a state-of-the-art model. To address this problem, three classifiers were used: MobileNetV3 Small, VGG-19, and FER-VT. The experimental results revealed that each model has distinct strengths and weaknesses, with MobileNetV3 Small being the most efficient for resource-constrained environments, VGG-19 achieving the highest accuracy across metrics while maintaining the ideal balance of performance and computational requirements, and FER-VT struggling with generalization issues. These findings emphasize the importance of choosing an appropriate model based on the specific application requirements, while balancing computational constraints and performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

From Constricted Models to State-of-the-Art for Facial Expression Classification

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月15日

查看全文
  • Ana Sofia Rodrigues,
  • Júlio Castro Lopes,
  • Rui Pedro Lopes

摘要

Facial expressions are an important channel for interpersonal communication and comprehension since people externalize their emotions through a variety of facial expressions. Technology, in particular, deep learning algorithms, can detect and analyze human emotions in real-time, which paves the way for advanced user interfaces or adjustable devices and applications. Based on this, the work described in this paper presents a system that identifies three groups of emotions, positive, negative, and neutral, in three execution scenarios: in resource-limited devices, such as mobile phones, for desktop or web applications, and with a state-of-the-art model. To address this problem, three classifiers were used: MobileNetV3 Small, VGG-19, and FER-VT. The experimental results revealed that each model has distinct strengths and weaknesses, with MobileNetV3 Small being the most efficient for resource-constrained environments, VGG-19 achieving the highest accuracy across metrics while maintaining the ideal balance of performance and computational requirements, and FER-VT struggling with generalization issues. These findings emphasize the importance of choosing an appropriate model based on the specific application requirements, while balancing computational constraints and performance.