Embedded Files
Disclaimer: Please note that the talk titles and abstracts listed below are provisional. They are subject to change as we finalize the workshop program.
Professor at the Technical University of Munich
Title of Talk: Building Interactable 3D Spaces from Imperfect Observations
Abstract: TBA
Professor at the University of Oxford
Title of Talk: Building a 3D Foundation for Spatial AI
Abstract: I will discuss our work on developing a 3D foundation for Spatial AI, which I see as a prerequisite for understanding, acting, and creating in the physical world. I will show that general-purpose transformers, such as VGGT, excel at 3D reconstruction and can be extended to new-view synthesis. I will also explain how Dynamic Point Maps make it easy to further generalise these models to 4D reconstruction. Next, I will demonstrate that the performance of these models is far from saturated, with significant gains achieved by increasing the training data fifteen-fold in the new VGGT-Ω. I will argue that, as 3D transformers mature, they provide a credible basis for integration into multimodal systems, including Vision–Language–Action models. Finally, beyond analysis, I will showcase recent progress in 3D generation, including fully articulated and simulation-ready 3D objects.
Latest reserach project: VGGT-Ω
Professor at the University of Tübingen
Title of Talk: TBA
Abstract: TBA
Page updated
Google Sites
Report abuse