DAiSEE: Towards User Engagement Recognition in the Wild

Recognizing user engagement plays an important role in several contemporary vision applications including advertising, healthcare, and e-learning. However, the lack of any publicly available dataset to address this problem severely limits the development of methodologies that can help make progress to address this challenge. In this paper, we introduce DAiSEE, a free-to-use large dataset comprising of 8000 video snippets captured from 95 users for recognizing user engagement in the wild. Baseline results using standard feature extraction and classification methods show the difficulty of working with this dataset. We believe that DAiSEE would provide the research community with challenges in feature extraction, context-based inference as well as the development of suitable machine learning methods that can consider annotator statistics in the training process itself, thus providing a springboard for further research.