Xuyi Meng

I am a 2nd-year CS M.S. student at University of Pennsylvania, advised by Prof. Lingjie Liu. Before that, I have acquired my B.E. degree in CS with highest distinction from Nanyang Technological University, Singapore, where I had the great opportunity to work in MMLab@NTU, advised by Prof. Chen Change Loy and Prof. Bo Dai. During my undergrad, I also spent a great summer in 2022 visiting UCSD and working with Prof. Hao Su.

Email  /  CV  /  Github

My research focuses on high-quality 3D generation by leveraging abundant 2D data and the powerful priors of large foundation models through differentiable neural rendering. I am also dedicated to detailed human generation and reconstruction, with experience in real-time renderable human avatar reconstruction from video and generating 3D-aware, high-fidelity human faces. My objective is to develop models that maintain consistency and adhere to physical laws in 3D/4D spaces while learning from rich 2D data through differentiable rendering.

I am actively looking for a Ph.D. position in 2025 Fall!

We reframe the challenging task of direct 3D generation within a 2D diffusion framework, by decomposing the splatter image into a set of attribute images. The use of pretrained 2D diffusion model enables high-quality 3D generation and good generalization ability to in-the-wild data.

LN3Diff creates high-quality 3D object mesh from text within 8 SECONDS.

We propose E3DGE, an encoder-based 3D GAN inversion framework that yields high-quality shape and texture reconstruction.

We present SPNR, which encodes sparse point clouds into neural volume representations to produce images with high visual quality. Given sparsely sampled surface points, SPNR generates disentangled density and color volumes and utilizes volumetric rendering to produce viewconsistent high-quality images.

Thanks for the amazing source code from Jon Barron's website 🧚