Xiaolong Wang
University of California-San Diego, CA·Computer Science·Early-career
For a household robot to help someone, or for a doctor to review how an elderly patient moves around at home, a computer must understand how people handle objects in three dimensions and over time. Most current computer vision systems understand flat 2D images.
- Fields
Computer Science
Engineering
- Topics
computer vision
human-object interaction
4D perception
- Grants
- 1 known grant, through Apr 2028
What they work on
- Recognize objects in videos of everyday environments without a fixed category list, training with text supervision and self-supervision instead of detailed labels.
- Reconstruct the 3D structure of humans and objects from video using minimal 3D supervision combined with manually defined priors.
- Reason about contact and interaction between people and objects over time using a 4D Interaction Graph that links them across space and time.
- Develop a 4D Vision Education Platform with course materials on generalization in computer vision and robotics applications, for university courses and K-12 tutorials.
More about the research
They work in labs or tidy scenes because they depend on detailed 3D labels that real homes and streets do not provide.
Why it matters
Computers that understand how people physically interact with objects in ordinary settings could help household robots predict and assist with daily tasks. They could also help doctors analyze the behavior of elderly people to improve their care. The project names social science as another field that could benefit.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Shuran Song
Stanford University, CA·Computer Science·Early-career
Develop hierarchical task representations, built in hyperbolic space (a tree-like curved geometry), learned from large-scale unlabeled instructional videos to model the steps of everyday activities such as cooking and repair.
Carl Vondrick
Columbia University, NY·Computer Science·Early-career
Build open-world recognition models for satellite, aerial and ground imagery from vision-language models without language annotations, using cross-modal contrastive learning and reasoning about change over time.
Changhyun Choi
University of Minnesota, MN·Computer Science·Early-career
Develop context-aware grasping algorithms that let robots consider clutter, reachability and collisions when picking objects from crowded scenes, using visual scene understanding and self-supervised learning.
The school
University of California-San Diego: 84 labs on PI Pages, $1.9B a year on research, 637 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.