Class-attention video transformer for engagement prediction
摘要
In this paper, we propose the Class Attention in Video Transformer (CavT), an end-to-end method designed to process both long and short variant-length videos for student engagement prediction. CavT introduces a single vector for class embedding and incorporates the Binary-Order Representatives Sampling (BorS) technique to augment the dataset by adding multiple video sequences. Our method outperforms the state-of-the-art with MSE values of 0.0495 on the EmotiW-EP and 0.0377 on the DAiSEE datasets, providing a robust and scalable solution for engagement prediction.