Yihan Wang (王羿涵)

I'm building physical intelligence at a stealth startup. I did research in geometric vision during my Ph.D. at PrincetonVL and my internship with Meta’s Surreal team. Previously, I received my B.Eng. from Tsinghua University (Yao Class).

Here are some of the projects I led:

WAFT-Stereo (ECCV ’26): We built an efficient model that reconstructs metric 3D from high-resolution stereo image pairs^{Two images of the same scene from slightly different viewpoints, like our two eyes. Comparing them helps estimate depth.} at sub-pixel accuracy. The full training recipe is open source, and the model has been integrated into Meta XR, Project Aria, and AMETEK Creaform.
WAFT (ICLR ’26, Oral): We built a simpler and more accurate optical flow^{Optical flow describes the apparent 2D motion of pixels between consecutive video frames. Arrows show the direction and distance of that motion.} estimator using high-resolution feature warping. The model has been integrated into Adobe After Effects. Learn more here.
SEA-RAFT (ECCV ’24, Best Paper Finalist): We built a highly efficient optical flow estimator with strong generalization. The model has been used in iPhone’s camera image signal processing (ISP) pipeline.