Objaverse-XL: A Universe of 10M+ 3D Objects
This addresses the data bottleneck for 3D vision tasks, enabling broader progress in the field, though it is incremental as it scales up existing data collection efforts.
The authors tackled the lack of large-scale 3D data by creating Objaverse-XL, a dataset of over 10 million 3D objects, and demonstrated its utility by training Zero123 on novel view synthesis with over 100 million rendered images to achieve strong zero-shot generalization.
Natural language processing and 2D vision models have attained remarkable proficiency on many tasks primarily by escalating the scale of training data. However, 3D vision tasks have not seen the same progress, in part due to the challenges of acquiring high-quality 3D data. In this work, we present Objaverse-XL, a dataset of over 10 million 3D objects. Our dataset comprises deduplicated 3D objects from a diverse set of sources, including manually designed objects, photogrammetry scans of landmarks and everyday items, and professional scans of historic and antique artifacts. Representing the largest scale and diversity in the realm of 3D datasets, Objaverse-XL enables significant new possibilities for 3D vision. Our experiments demonstrate the improvements enabled with the scale provided by Objaverse-XL. We show that by training Zero123 on novel view synthesis, utilizing over 100 million multi-view rendered images, we achieve strong zero-shot generalization abilities. We hope that releasing Objaverse-XL will enable further innovations in the field of 3D vision at scale.