Jeff Tan

Hello! I'm a second-year MS in Robotics student at Carnegie Mellon University, advised by Prof. Deva Ramanan and often collaborating with Prof. Shubham Tulsiani. Previously, I was an undergrad in CS at CMU. I am grateful for support from the NSF Graduate Research Fellowship.

In the future, I am interested in building generalist robots that can achieve human-level intelligence, dexterity, and agility.

CV  /  Github  /  Email  /  Twitter

profile photo
Research

DiffusionSfM: Predicting Structure and Motion via Ray Origin and Endpoint Diffusion
Qitao Zhao, Amy Lin, Jeff Tan, Jason Y. Zhang, Deva Ramanan, Shubham Tulsiani
CVPR 2025
Website | Paper

From a set of multi-view images, we learn a denoising diffusion model that outputs scene geometry and cameras in global frame.

DressRecon: Freeform 4D Human Reconstruction from Monocular Video
Jeff Tan, Donglai Xiang, Shubham Tulsiani, Deva Ramanan, Gengshan Yang
3DV 2025 (Oral)
Website | arXiv | Github

From a single monocular video, we reconstruct humans in loose clothing and interacting with objects, using a hierarchical deformation field and image-based priors.

Distilling Neural Fields for Real-Time Articulated Shape Reconstruction
Jeff Tan, Gengshan Yang, Deva Ramanan
CVPR 2023
Website | Paper | Github

We learn real-time feed-forward pose and shape predictors, by distilling knowledge from offline differentiable rendering optimizers.

Projects

Lab4D
Gengshan Yang, Jeff Tan, Alex Lyons, Neehar Peri, Deva Ramanan
Website | Github

A Python library for 4D reconstruction of humans, animals, and scenes from monocular videos.

Natural Dexterous Piano Playing at Scale With Video Hand Priors
Jeff Tan, Yuanhao Wang, Haoyang He
Report

We control dexterous simulated robot hands to play piano, using Internet videos of human pianist demonstrations.

Cleaning Casually Captured Splatting Scenes with Diffusion Priors
Jeff Tan, Bhuvan Jhamb, Joel Julin, Roshan Roy
Report

We fine-tune image-conditioned diffusion models to simultaneously remove ghostly artifacts and infill plausible geometry at novel views.

Physically Based Renderer
Jeff Tan
Report

A physics-based renderer for photorealistic images that supports Monte Carlo path tracing with physically realistic materials, bidirectional path tracing, and volume rendering.

C0 Compiler
Jeff Tan, Rachel Yuan
Report

A compiler for a type-safe subset of C, outperforming gcc -O1 by 36.9% on average on CMU 15-411's benchmark suite.

Misc
CMU SCS Logo Teaching Assistant, Physics-Based Rendering (15-468): S23, S24
Teaching Assistant, Parallel Computation (15-418): F21, S22, S23
Teaching Assistant, Introduction to Computer Systems (15-213): F21

Credits to Jon Barron for this website's template.