Humphrey Shi
Humphrey Shi is a computer vision researcher whose work spans multimodal AI, image segmentation, and video generation. He earned an engineering physics degree at Tsinghua in 2009 and master's and doctoral degrees in electrical and computer engineering at UIUC in 2016 and 2017, then worked at IBM and became an assistant professor at Oregon. His work includes OneFormer, a universal image segmentation framework, and Text2Video-Zero, which adapts text-to-image diffusion models for video generation. He founded Picsart AI Research, led a team developing AI for creative communication, received the NSF CAREER Award in 2023, and joined Georgia Tech as an associate professor. Since November 2025, he is NVIDIA's vice president of High-Performance AI and a Georgia Tech associate professor.
- Now at
- Georgia Institute of Technology and NVIDIA
Timeline
-
2005–2017
Engineering physics and an ECE doctorate
He earns a B.E. in engineering physics at Tsinghua University in 2009 and M.S. and Ph.D. degrees in electrical and computer engineering at UIUC in 2016 and 2017.
-
November 2017–December 2019
Computer vision at IBM
From November 2017 to December 2019, he works as a research staff member in computer vision at IBM's T. J. Watson Research Center.
-
2019–2023
Assistant professor at Oregon
From 2019 to 2023, he is an assistant professor of computer science at the University of Oregon. His research focuses on multimodal AI that combines images, video, audio, geometry, and text.
-
February 2021–April 2025
Chief scientist at Picsart
From February 2021 to April 2025, he serves as chief scientist at Picsart and founder of Picsart AI Research. He leads scientists, engineers, and creatives developing AI for creative communication.
-
2022
OneFormer unifies segmentation
In 2022, he coauthors OneFormer, a universal image segmentation framework that unifies semantic, instance, and panoptic segmentation in a multi-task train-once design.
-
2023
Text2Video-Zero for video generation
In 2023, he coauthors Text2Video-Zero, a low-cost approach that uses existing text-to-image diffusion methods for video generation without added training or optimization.
-
2023
A Georgia Tech professorship
In 2023, he joins Georgia Tech's School of Interactive Computing as an associate professor after receiving the NSF CAREER Award in June 2023.
-
Since November 2025
Vice president of AI at NVIDIA
In November 2025, he becomes NVIDIA's vice president of High-Performance AI and continues as an associate professor in Georgia Tech's School of Interactive Computing. As of October 2026, he leads work on high-performance, multimodal, and agentic AI across models, kernels, compilers, and hardware.
Career
- Since 2025NVIDIAVice President of High-Performance AI
- Since 2023Georgia Institute of TechnologyAssociate Professor
- 2021–2025PicsartChief Scientist, Founder
- 2019–2023University of OregonAssistant Professor
- 2017–2019IBM ResearchResearch Staff Member
Education
- 2017University of Illinois Urbana-ChampaignPh.D.
- 2016University of Illinois Urbana-ChampaignM.S.
- 2009Tsinghua UniversityB.E.
Sources and credits
Photo: We have no published photo of Humphrey Shi yet. Send us a free-licensed photo.
- Humphrey Shi | Electrical & Computer Engineering | Illinois
- Incoming Associate Professor Takes Holistic View of AI and Computer Vision | College of Computing
- Humphrey Shi | School of Interactive Computing
- OneFormer: One Transformer to Rule Universal Image Segmentation
- Text2Video-Zero: Text-to-Image Diffusion Models are Zero-Shot Video Generators
- Author: Humphrey Shi | NVIDIA Technical Blog



