Jun Gao

Jun Gao is a computer vision and graphics researcher who works on differentiable rendering, 3D generation and video generation. After internships at Microsoft Research Asia and NVIDIA, he became an NVIDIA research scientist in 2019 and earned a Ph.D. at the University of Toronto under Sanja Fidler. His coauthored papers include DIB-R in 2019, Deep Marching Tetrahedra in 2021, GET3D and Magic3D in 2022, and Difix3D+ in 2025, a CVPR 2025 Best Paper Award Candidate. A paper he coauthored won the Best Paper Award at SIGGRAPH Asia 2023, and MOTIVE, which traces motion in video generation back to training clips, received an ICML 2026 Outstanding Paper Honorable Mention. He is an assistant professor in the EECS department at the University of Michigan.

Now at: NVIDIA and University of Michigan
Badges: Bridge

Timeline

2017–2019

Internships and the DIB-R renderer

He interns at Microsoft Research Asia from September 2017 to June 2018 and at NVIDIA from October 2018 to November 2019. In 2019, he coauthors DIB-R with Wenzheng Chen, Sanja Fidler and others, a differentiable rendering framework that computes gradients analytically for all pixels in an image.
Since 2019

Research scientist at NVIDIA

In November 2019, he becomes a research scientist at NVIDIA. At the University of Toronto, he is a Ph.D. candidate advised by Sanja Fidler in March 2024 and goes on to receive the degree.
2021

Deep Marching Tetrahedra

In 2021, he coauthors Deep Marching Tetrahedra (DMTet) with Tianchang Shen, Kangxue Yin, Ming-Yu Liu and Sanja Fidler. DMTet is a deep 3D conditional generative model that synthesizes high-resolution 3D shapes from simple user guides such as coarse voxels.
2022

GET3D and Magic3D

In 2022, he is first author of GET3D, a generative model that directly generates explicit textured 3D meshes with complex topology and learns from 2D image collections. He also coauthors Magic3D with Chen-Hsuan Lin, Ming-Yu Liu, Sanja Fidler and others, a text-to-3D method that creates high-quality 3D mesh models in 40 minutes.
2023

A SIGGRAPH Asia best paper

In 2023, Adaptive Shells for Efficient Neural Radiance Field Rendering, a paper he coauthors with Zian Wang, Tianchang Shen, Sanja Fidler and others, wins the Best Paper Award at SIGGRAPH Asia. The method renders scenes up to 10 times faster than older methods.
2025

Difix3D+ at CVPR 2025

In 2025, he coauthors Difix3D+ with Jay Zhangjie Wu, Sanja Fidler, Zan Gojcic, Huan Ling and others, a pipeline that uses single-step diffusion models to improve 3D reconstruction and novel-view synthesis. The paper is a CVPR 2025 Best Paper Award Candidate.
2026

MOTIVE at ICML 2026

In 2026, MOTIVE (Motion Attribution for Video Generation), coauthored with Xindi Wu, Sanja Fidler, Antonio Torralba and others, receives an Outstanding Paper Honorable Mention at ICML 2026. The framework identifies which training clips affect motion in text-to-video models and guides fine-tuning data selection, reaching a 74.1% human preference win rate against the pretrained base model.
As of October 2026

Assistant professor at Michigan

Introduced in January 2025 as an incoming assistant professor at the University of Michigan, he is an assistant professor in its EECS department. His research aims at generative world models that are controllable, consistent, interpretable and scalable, for realistic interactions with humans and autonomous agents.