ARCTIC: A Dataset for Dexterous Bimanual Hand-Object Manipulation

A motion-capture dataset of two-handed dexterous manipulation of articulated objects, with precise 3D meshes for hands and objects every frame.

ARCTIC was released in 2022 by ETH Zurich, the Max Planck Institute for Intelligent Systems, and the University of Amsterdam, published at CVPR 2023. Ten subjects used both hands to manipulate 11 articulated objects (objects with moving parts, such as scissors and a laptop), producing 339 sequences and 2.1 million frames of imagery, filmed synchronously from 8 fixed viewpoints and 1 egocentric viewpoint. The team used a 54-camera infrared Vicon optical motion-capture system to track small markers attached to the hands and objects, and solved for a MANO hand mesh, an SMPL-X body mesh, object pose and joint angle, and hand-object contact information for every frame. Earlier hand-object interaction datasets were mostly single-handed, rigid-object, and slow-motion; ARCTIC fills the gap with precise 3D ground truth for fast, two-handed manipulation of moving objects, and can also serve as a reference for learning dexterous manipulation from human data.

ExampleARCTIC contains sequences of both hands opening a laptop and opening and closing scissors, with the 3D mesh of both hands and the rotation angle of the object's moving part given for every frame.

Related

Hand-Object Interaction · MANO · SMPL · Optical Motion Capture · Articulated Object · HOI4D

Sources

ARCTIC: A Dataset for Dexterous Bimanual Hand-Object Manipulation (arXiv 2204.13662)
ARCTIC 项目主页 (Chinese)

As of

2023-06