Minh Hoai Nguyen – Research

Research Interests

I am interested in developing algorithms for human behavior analysis. By human behavior, I mean the actions or activities of a person or a group of people recorded in videos. Human behavior also refers to the interaction between hands and other objects, and sometimes the behavior of a person can be determined based only on the observation of their hands. Human behavior could also include facial or head movements, as there might be a need to quantify how happy or depressed a person is. Human behavior also includes attentional behavior, and I am interested in discovering the parts of an image or a video that attract visual attention, as well as predicting and manipulating visual attention. As for the analysis tasks, I am working on both recognition and prediction tasks. Recognition is concerned with the past, while prediction refers to the future. There is a third category of analysis tasks called early recognition which I am particularly interested in. Early recognition refers to the present, and the task is to detect and categorize an ongoing activity.

Toward better algorithms for human behavior analysis, I am also tackling other computer vision problems, including hand detection and human counting. My research also extends to machine learning because human behavior is so complex that algorithms for human behavior analysis need to be trained rather than hand-designed. In general, the performance of a learning-based method depends on the quantity and quality of human annotated data to train it, but providing detailed annotations for human behavior is a laborious and subjective process. Thus underlying many of my research projects is the development of algorithms that can learn from weakly-labeled, noisily-labeled, or unlabeled data. These algorithms are motivated by specific application scenarios, but they often have broader impacts beyond the intended applications.

Selected Current and Past Projects

  • Visual Counting

  • Deep Learning for Material Science