Yuliang Xiu

Despite advances in MLLMs, they still struggle with dynamic reasoning, falling far short of human capabilities in interpreting continuous visual cues.

Forget global coordinates: this new method unlocks long-context and streaming 3D reconstruction by predicting relative constraints between frames.

Fix freaky AI-generated hands and other anatomical nightmares with a new preference learning method that surgically corrects anatomical errors in diffusion models.

Achieve millimeter-level accuracy in 3D human body fitting from multi-modal inputs, even with scale distortion common in AI-generated assets.

Animating 4D shapes just got easier: GaussiAnimate's "Skelebones" can reanimate unseen poses with 17% better PSNR than standard methods.

Achieve robust human body fitting across diverse clothing, poses, and input completeness by disentangling "undressing" and "dense fitting" stages.