PlatformMarketplaceSolutionsResourcesOpen DatasetsCommunityCompany
Owned byDatawhale
Jan 12, 2022 3:44 PM


Hand pose estimation in 2D images is a burgeoning research line that is booming in the field of computer vision. This problem is not only about the location of the hand itself, but also the localization of the hand joints,namely, the pose of the hand. The usefulness of this identification is wideranging, from its use for sign language recognition[1, 2] to the identification of more complex behaviors such as hand gestures. It is also worth highlighting its importance in virtual reality and augmented reality applications. In order to interact in a natural way in these virtual worlds, the hand pose detection problem must first be tackled. Furthermore, to achieve a seamlessly integrated experience in virtual worlds, non-intrusive methods must be utilized for the hand pose detection.


The proposed dataset is intended to be used for 3D and 2D hand pose estimation and hand area location. The dataset is structured in various sequences and each sequence is composed of a set of frames. A frame is a collection of different kinds of ground truth data for a certain instant in time. Th ground truth data provided includes:

  • A set of 3D points of the joints in the hand as provided by the Leap

Motion, without further processing.

  • Four color images of a hand as captured for each color camera.
  • Four sets of 2D points as the resultant projection of the 3D points to

each color camera coordinate frame.

  • Four bounding boxes computed from the projection of the 3D points

to each camera coordinate frame.


Francisco Gomez-Donoso, Sergio Orts-Escolano, and Miguel Cazorla. "Large-scale Multiview 3D Hand Pose Dataset". ArXiv e-prints 1707.03742, July 2017.

Data Preview
List Dataset Files
🎉Many thanks to Datawhale for contributing the dataset
Basic Information
Application ScenariosOthers
TasksPose Estimation
LicenseNot Available
Updated on2022-01-12 15:44:38
Data TypeImage
Data Volume49,062
Annotation Amount49,062
File Size769.89MB
Copyright Owner
Francisco Gomez-Donoso, Sergio Orts-Escolano, and Miguel Cazorla.
Francisco Gomez-Donoso, Sergio Orts-Escolano, and Miguel Cazorla.