MANO

The most widely used parametric 3D hand model, describing a hand with a small number of shape and pose parameters.

MANO, short for hand Model with Articulated and Non-rigid defOrmations, was proposed by Javier Romero and Dimitrios Tzionas in Michael Black's group at the Max Planck Institute for Intelligent Systems in Germany, published at SIGGRAPH Asia 2017. It was learned from about 1,000 high-precision 3D hand scans of 31 people: given shape parameters (commonly 10-dimensional, describing the hand's size and thickness) and pose parameters (the rotation of each finger joint, which can be compressed to fewer dimensions with principal component analysis), it outputs a 3D hand mesh complete with finger bending and skin deformation. It can also attach to the SMPL body model to form SMPL+H. In embodied AI, MANO is a common representation for learning dexterous manipulation from human-hand video: hand-reconstruction methods like HaMeR output MANO parameters, which are then mapped, through motion retargeting, onto a robot dexterous hand's joints.

ExampleMANO parameters are estimated frame by frame from a first-person video of a person picking up a cup, yielding a trajectory of finger joint angles, which is then retargeted into a demonstration for a robot dexterous hand.

Also called

hand Model with Articulated and Non-rigid defOrmations

Related

SMPL · HaMeR · Hand Pose Estimation · Motion Retargeting · Dexterous Hand · Human Video Data

Sources

MANO 官网（MPI-IS） (Chinese)
arXiv 2201.02610: Embodied Hands: Modeling and Capturing Hands and Bodies Together
GitHub: hassony2/manopth（MANO 的 PyTorch 实现） (Chinese)

As of

2017-11