A Quest for Building a Foundation 3D Digital Human Model
DescriptionDigital humans (DH) technology has a myriad of applications, but currently each DH application needs to train a task-specific model from scratch using a small domain-specific dataset (e.g., hair, body, face), which is highly inefficient and not scalable. In this talk, we will describe our unique approach to solving this task by training a unified and highly photorealistic 3D foundation model for DH from a large collection of in-the-wild 2D data that can be used to build many different downstream applications. We will describe our efforts so far on building a 3D DH foundation model and its implications for the future of DH research and industry.
Speaker
Event Type
Featured Session
TimeTuesday, 3 December 20242:15pm - 3:15pm JST
LocationHall B7 (2), B Block, Level 7
