Runfeng Li

Hi, I'm a research assistant at Brown University, working with Prof. James Tompkin, and collaborating with Prof. Matthew O'Toole and Dr. Christian Richardt. I did my master's in computer science at Brown and my undergrad in mathematics and computer science at UIUC.

Email: runfeng_li [at] brown [dot] edu

Github  /  Google Scholar  /  CV  /  Linkedin

Research

I'm interested in sensing, optimization/learning, and maybe some human-centered applications and more. Still exploring :)

I've worked around revealing good-looking images from partially corrupted measurements in physics-based ways. Specifically, dynamic scene modeling, physics-based reconstruction, radiance fields, and time-of-flight imaging.

We analyze ToF radiance field reconstruction and demonstrate multi-view scene unwrapping and single-view semi-transparent surface reconstruction with multi-frequency C-ToF.

We show that radiance field reconstruction from single-frequency continuous-wave raw time-of-flight images is fundamentally ill-posed. However, we find optimization biases in the 3D Gaussian parameterized radiance fields that help recover the most plausible geometry, which our model applies to dynamic scenes like fast-swinging baseball bats with quality comparable to or better than prior neural-field-based methods while retaining the efficiency of Gaussians.

We benchmark dynamic Gaussian splatting methods for monocular view synthesis, combining existing datasets and a new synthetic dataset to provide standardized comparisons and identify key factors affecting efficiency and quality.

Other Past Interests

We estimate Young's modulus by backpropagating video-reconstruction gradients through our Taichi implementation of PhysGaussian.

Reference: Jon Barron's Template.