Unveiling Emotions: Bridging the Gap in Children's Emotional Recognition
Arjun Malleswaran, Anthony Li, Hovhannes Broyan
amallesw@ucsd.edu, all010@ucsd.edu, hbroyan@ucsd.edu
Mentor: Tauhidir Rahman (trahman@ucsd.edu)
What is the Problem?
Emotional recognition has continued to be an evolving task within the deep learning space. However, a significant pitfall hampers our understanding and development of these tasks: the nuanced emotions of children. Traditional models, while effective with adult faces, stumble when faced with the diverse expressions of children. These models fail to capture the subtleties and intensities unique to children’s facial structure. This oversight limits the scope of emotional recognition tasks and impacts its applicability in critical areas such as early diagnosis of psychological disorders, where understanding a child's emotional state is crucial.
Current Solutions and Their Limitations
Current solutions have looked into children’s emotional responses to various reactions to psychological tasks and games. While facial expression recognition models have been used to assess these emotional responses quickly, their effectiveness is limited due to the models being predominantly trained on adult faces. Children display a wider range of emotions than adults, making children's emotion detection a unique challenge.
Proposed Solutions
Our project aims to tackle fine-tuning existing emotional recognition architectures and create a model that generalizes well to children's facial expressions. By leveraging techniques such as data augmentation and transfer learning from various pre-trained models, we aim to develop a model that thrives under real-world conditions.