Scaling 3D Digital Humans
In this talk, we explore the paradigm shift of bringing 3D digital human modeling into the era of large-scale foundation models. While 2D generative models have seen explosive growth, 3D human synthesis has often remained constrained by data scarcity and computational overhead. We bridge this gap by investigating the pre-training and post-training regimes for 3D avatars. We present findings demonstrating that pre-training on massive, diverse datasets—followed by targeted fine-tuning on high-fidelity, curated "clean" data—unlocks unprecedented generalization. This approach enables robust inference on out-of-distribution data, far surpassing traditional methods.
- Keynote
Speaker
Shunsuke Saito
Research Scientist, Meta Reality Labs Research
Shunsuke Saito is a Research Scientist at Meta Reality Labs Research in Pittsburgh, where he leads the effort on next generation digital humans. He obtained his PhD degree at the University of Southern California. His research lies in the intersection of computer graphics, computer vision and machine learning, especially centered around digital human, 3D reconstruction, and performance capture. His work has been published in SIGGRAPH, SIGGRAPH Asia, NeurIPS, ICLR, ECCV, ICCV and CVPR, three of w... read more