Yuanbo Yang

I am a Master's student at Zhejiang University, working with Prof. Yiyi Liao. I am also lucky to have collaboration with Prof. Andreas Geiger. Previously I obtained my B.Eng. degree in Electronics Engineering from Hangzhou Dianzi University in 2022.

My research interest lies in 3D Computer Vision and Generative Models, including:

• Structured and Controllable 3D/Video Generative Model
• Unleash the Power of “Free Lunch” (Synthetic Data, Pretrained Prior, etc.) to Enhance Generative Models
• Learning Structured Generative Representation from In the Wild Data

Email  /  CV  /  X  /  Github

Prometheus introduces a novel method for feed-forward scene-level 3D generation in seconds. Its key idea is to harness the power of pre-trained 2D priors to enable generalizable and efficient 3D synthesis – hence its name, Prometheus🔥.

UrbanGen proposes a solution for the challenging task of generating 3D urban radiance fields with photorealistic rendering, accurate geometry, high controllability, and diverse city styles.

ChronoDepth addresses the challenge of streamed video depth estimation with video diffusion model.

TeFF targets learning 3D-aware GANs from unposed images, for which we propose to perform on-the-fly pose estimation of training images with a learned template feature field (TeFF).

MaPa introduces a method to create segment-wise procedural material graphs as the appearance representation, which supports high-quality rendering and provides significant flexibility in editing.

UrbanGIRAFFE leverages a coarse 3D panoptic prior to guide a 3D-aware generative model, enabling photorealistic 3D-aware image synthesis with diverse controllability, including large camera movement, stuff editing, and object manipulation.

Last updated Dec. 2024
Thanks Dr. Jon Barron for sharing the source code of his personal page.