Cewu Lu

Building the foundation for computer vision, his breakthrough research findings achieve deeper level visual semantic understanding.

After joining Professor Fei-Fei Li’s lab at Stanford, Cewu Lu technically drives a research topic in “Visual Relationship Detection," looking to study the relationship between objects in an image. The results of his research would become important tools in developing high-level semantic understanding. In other words, Cewu was building the foundation for computer vision knowledge construction.

During his research, Cewu proposed a novel solution by embedding language knowledge into a deep learning model, which allowed knowledge of visual relationships with large samples to be transferred to the one that has small training samples. It can even infer unseen visual relationships.

Cewu joined the Geometric Computing Lab at Stanford later on, focusing his research on rich semantic expression. For high level visual tasks such as human-object interaction, object functionality and affordance, and robotic control, he proposed a new tool named as “part-state” to explore rich semantic expression techniques.

Now, as a Professor at Shanghai Jiao Tong University, Cewu built a machine vision lab and achieved new breakthroughs in human activity understanding and robotics vision. His team develops a human pose understanding system – Alphapose outperforms the state-of-the-art system by 8.2%. Currently, he is working on jointly learning vision and haptic together for robotics systems with new theoretical framework. Guided by human-object interaction knowledge, learnt system is expected to ensure robot operate automatically like humans in complex scenarios.