Google Scanned Objects

Google's open-source library of high-precision 3D scans of 1,030 real household objects.

Google Scanned Objects (GSO) is a library of 3D object assets that Google Research released in 2022, containing 1,030 real household items — shoes, toys, tableware, packaging boxes, and more. Google built a dedicated scanning rig that uses structured light (projecting a pattern onto the object and inferring its shape from camera images) combined with DSLR photography for high-dynamic-range color, producing mesh models with realistic texture. The models are pre-processed into formats that load directly into Gazebo and PyBullet, hosted on Gazebo Fuel, and released under the CC-BY 4.0 license. It solves the problem of simulators having too few objects, and objects that look too fake: grasping, placement, and synthetic perception data all need large numbers of objects that resemble real things in both shape and appearance, and GSO is one of the commonly used ready-made sources for this, often combined with asset libraries like YCB and Objaverse.

ExampleShoe and toy models from GSO are downloaded from Gazebo Fuel and scattered randomly across a PyBullet tabletop scene to generate grasping training data in bulk.

Also called

GSO, Scanned Objects by Google Research

Related

Simulation Assets · YCB Object and Model Set · Objaverse · ShapeNet · Synthetic Data · Gazebo

Sources

Google Scanned Objects: A High-Quality Dataset of 3D Scanned Household Items (arXiv)
Scanned Objects by Google Research: A Dataset of 3D-Scanned Common Household Items (Google Research Blog)

As of

2022-06