Human-computer Interaction Framework project
Current state-of-the-art human-computer interaction platform    still focuses on voice/text-based interaction. The goal of this      project to build the next-generation human-computer interaction  framework that empowers visually grounded interactions for        physical robots or virual agents.

Visual supervision evaluation platform
Current computer vision (CV) systems are evalauted in controlled  environment such as object detection or segmentation. The goal  of this project is to create a benchmark that evalaute systems'     robustness and accuracy in real-world scenairos, which in turn    foster future research for real-world CV systems.

Multimodal public opinion understanding algorithm
Multimodal understanding is becoming increasingly important for  social media trend analyisis. The goal of thi sproject to develop    algorithms and computation models that better parse and            predict event trends and build event knowledge base.