Research
Building the ImageNet for 3D
We are building one of the world's largest collections of reality-grade 3D digital twins to improve the quality of computer vision, generative Al, robotics, and Physical Al models. Our dataset is designed to provide high-quality, physically accurate 3D assets at unprecedented scale, enabling the next generation of Al models.
In collaboration with our research partners, we are developing large-scale research datasets and creating SimReady digital twins to support benchmarking in computer vision, robotics, and Physical Al research. Through this initiative, we aim to advance research on the Sim2Real gap - one of the most important challenges in Al today.
Research Partners
Current Research Areas
Our research is focused on solving some of the most important challenges in computer vision, generative Al, robotics, and physical Al. By combining large-scale, reality-grade 3D data with academic research, we develop the foundations for the next generation of Al models.
Relightable 3D
Reconstruction
+
Pushing the boundaries of 3D reconstruction at both the hardware and software levels to recover geometry, materials, and physical properties with the highest possible fidelity.
Benchmarking 3D
Data Quality
+
Measuring how data quality impacts model performance across computer vision, robotics, and generative AI.
SimReady Digital
Twins for Physical AI
+
Developing reality-grade 3D datasets for robotics and Physical AI in collaboration with leading AI research teams, including NVIDIA Research.
Generative 3D Al
+
Advancing the next generation of 3D generative AI through large-scale, reality-grade datasets.
The Digital Twin Catalog by Meta
We partnered with Meta to release the Digital Twin Catalog, the most photorealistic open-source 3D dataset to date. Our scanners powered the object capture pipeline behind the dataset.