SMPL

Abstract

We present a learned model of human body shape and pose-dependent shape variation that is more accurate than previous models and is compatible with existing graphics pipelines. Our Skinned Multi-Person Linear model (SMPL) is a skinned vertex-based model that accurately represents a wide variety of body shapes in natural human poses. The parameters of the model are learned from data including the rest pose template, blend weights, pose-dependent blend shapes, identity-dependent blend shapes, and a regressor from vertices to joint locations. Unlike previous models, the pose-dependent blend shapes are a linear function of the elements of the pose rotation matrices. This simple formulation enables training the entire model from a relatively large number of aligned 3D meshes of different people in different poses. We quantitatively evaluate variants of SMPL using linear or dual-quaternion blend skinning and show that both are more accurate than a Blend-SCAPE model trained on the same data. We also extend SMPL to realistically model dynamic soft-tissue deformations. Because it is based on blend skinning, SMPL is compatible with existing rendering engines and we make it available for research purposes.

Surface based motion retargeting by preserving spatial relationship

“…Given an arbitrary target calibration shape, we may have two possible solutions: (1) We deform the given target calibration pose such that it has a similar pose as source using shape deformation methods such as [Ben-Chen et al 2009a;Lipman et al 2008; Wang (2) We embed the deformation into our retargeting framework and force the spatial relationship constraint to guide the synthesized target pose. While we have shown many retargeting examples both in the paper and supplemental video, we also want to compare our algorithm with the other methods such as the Skinned Multi-Person Linear model [Loper et al 2015] in the future.…”

Virtual Agents from 360° Video for Interactive Virtual Reality

“…Second, the output data must enable creating a textured, animated 3D character mesh. Methods that estimate parameters for the Skinned Multi-Person Linear (SMPL) model [12] are practical in this sense, especially for VR research, since the model is easy to understand and has been adapted for use with the Unity game engine. Notable approaches based on this model include those presented by Kanazawa et al [9], Güler et al [8], and Alldieck et al [1,2].…”

Motion Capture Research: 3D Human Pose Recovery Based on RGB Video Sequences

“…The deep convolutional neural network is used to solve the SMPL model [11] parameters by the obtained 2D human pose and the input human height (H). This section includes the SMPL model, the end-to-end network structure, and the model pre-training process.…”

“…, P 9K ] ∈ R 3N×9K . Through the method of grid alignment [33], the multi-pose dataset [11] and multi-shape dataset [34] obtained from 3D scanning are aligned with the grid of SMPL model, then the aligned dataset is used to train and solve the model parameters: Φ.…”

“…To achieve the aforementioned goal, this paper proposes an end-to-end solution combined with a deep convolutional neural network, which can restore a 3D human grid model from single image data, and then smooth the serialized 3D human model using the correlation between RGB video frames and obtain the final human grid model. The solution proposed is to calculate the parameters of the parameterized skinned multi-person linear (SMPL) model [11] to restore the 3D pose of the human body. To constrain the arbitrariness in the solution process: On the one hand, we use 2D/3D skeleton point constraints to obtain a more accurate human-body model and combine human height constraints to reduce the arbitrariness problem in the two-dimensional (2D) to 3D solution process.…”