Date
Publisher
arXiv
In early childhood education, accurately detecting collaborative and behavioral engagement is essential to foster meaningful learning experiences. This paper presents an AI driven approach that leverages Vision Transformers (ViTs) to automatically classify children s engagement using visual cues such as gaze direction, interaction, and peer collaboration. Utilizing the ChildPlay gaze dataset, our method is trained on annotated video segments to classify behavioral and collaborative engagement states (e.g., engaged, not engaged, collaborative, not collaborative). We evaluated six state of the art transformer models: Vision Transformer (ViT), Data efficient Image Transformer (DeiT), Swin Transformer, VitGaze, APVit and GazeTR. Among these, the Swin Transformer achieved the highest classification performance with an accuracy of 97.58 percent, demonstrating its effectiveness in modeling local and global attention. Our results highlight the potential of transformer based architectures for scalable, automated engagement analysis in real world educational settings.
Study at a glance
- Question: Can vision transformer models accurately classify children's behavioral and collaborative engagement from video images?
- Setting and data: 4,538 labeled image frames from 401 video clips of young children interacting with adults indoors (ChildPlay Gaze dataset).
- Method: Model evaluation and comparison of six vision transformer architectures on image classification.
Why it matters
Traditional engagement measurements are labor-intensive and coarse-grained. Automated computer vision can track nonverbal cues, but existing approaches struggle with class imbalance, scalability, and fine-grained detection of behavioral and collaborative participation in early childhood.
Key findings
- The Swin Transformer achieved the highest overall performance, reaching 97.58% accuracy, 0.9805 recall, and an F1-score of 0.9701.
- DeiT recorded the highest precision at 0.9623, while baseline Vision Transformer also performed strongly with an F1-score of 0.9583.
- GazeTR achieved 95.70% accuracy, substantially outperforming the lightweight VitGaze model which reached 85.02% accuracy.
Summary drafted with AI from the full paper and may contain errors. See the full paper for details.
What is the application?
Who age?
Why use AI?
Study design
