YCB Object and Model Set

A standard set of purchasable everyday objects with 3D scans, letting grasping and manipulation research compare results directly.

YCB was proposed in 2015 by researchers at Yale, Carnegie Mellon, and Berkeley (its name comes from the three schools' initials), with the paper published in IEEE Robotics & Automation Magazine. It includes everyday objects numbered 1 through 73, split into five categories — food, kitchen items, tools, shape primitives, and task objects — such as cans, mugs, a power drill, wooden blocks, and rope. Every object was scanned on a multi-camera turntable, yielding 600 RGB-D images, 600 high-resolution color photos, segmentation masks, and a textured 3D mesh. Before YCB, each lab picked its own objects and results couldn't be compared directly; YCB let everyone use the same physical objects and the same models in simulation, and came with template manipulation-benchmark protocols. Today it is commonly imported into simulators as grasping targets, and is also the object source behind several pose-estimation datasets.

ExamplePoseCNN's YCB-Video pose-estimation dataset (2017) was built by filming YCB objects on cluttered tabletops, totaling 133,827 frames.

Also called

YCB, YCB Objects, Yale-CMU-Berkeley Object Set

Related

Grasping · Simulation Assets · 6D Object Pose Estimation · Benchmark · Google Scanned Objects · BOP (Benchmark for 6D Object Pose Estimation)

Sources

Benchmarking in Manipulation Research: The YCB Object and Model Set and Benchmarking Protocols (arXiv 1502.03143)
PoseCNN: A Convolutional Neural Network for 6D Object Pose Estimation in Cluttered Scenes (arXiv 1711.00199)

As of

2015-02