Visual Modeling (VISMOD) group focuses on research of innovative machine learning methodologies and technologies, in order to solve real-world computer vision problems and AI+ problems, specifically human-centric vision scenarios including image and video representation, understanding and generation, as well as application of AI methods in other science and technology domains.

The current main research topics include:

●  Models and algorithms in machine learning, especially

1)    Learning under data distribution shift: long-tailed, few-shot, meta-learning, etc.

2)    Learning under weak annotations: unsupervised/semi-supervised/weakly supervised learning, etc.

3)    Anomaly and generalization dilemma: anomaly detection, out-of-distribution (OOD) detection/ generalization, etc.

●  Human-centric visual modeling

1)    Person re-identification

2)    Human pose, motion understanding and generalization

3)    Human-centric multimodal large language models

●  AI4Science

1)    Multimodal large language models with scientific data 

2)    Deep learning models in specific domains