SMPL-IKS: An Inverse Kinematic Solver for 3D Human Mesh Recovery

We present SMPL-IKS, an inverse kinematic solver to operate on the well-known Skinned Multi-Person Linear model (SMPL) to recover human mesh from 3D skeleton. The challenges of the task are threefold: (1) Shape Mismatching. (2) Error Accumulation. (3) Rotation Ambiguity. Instead of recovering human mesh from costly vertice up-sampling or iterative optimization as in previous methods, SMPL-IKS directly regresses the SMPL parameters (i.e., shape and pose parameters) in a clean and efficient way. Specifically, we propose to infer skeleton-to-mesh via two explicit mappings viz. SI and IK, the former maps bone length to shape parameters, and the latter maps bone direction to pose parameters. Moreover, we design two adaptive pose refinement mechanisms, termed PR, to alleviate the error accumulation problem. SMPL-IKS is general and thus extensible to MANO or SMPL-H. Extensive experiments are conducted on various benchmarks for body-only, hand-only, and body-hand scenarios. Our model surpasses the state-of-the-art methods by a large margin while being much more efficient. Data and code is available at https://github.com/Z-Z-J/SMPL-IKS

本报告介绍SMPL-IKS，一款基于知名的可变形多人体线性模型（SMPL）的逆运动学求解器，旨在从三维骨骼中恢复人体网格。该任务的挑战分为三个方面：（1）形状不匹配；（2）误差累积；（3）旋转模糊。与先前方法中通过昂贵的顶点上采样或迭代优化从人类网格中恢复相比，SMPL-IKS直接以干净高效的方式回归SMPL参数（即形状和姿态参数）。具体而言，我们提出通过两种显式的映射方式，即SI和IK，来推断骨骼到网格。其中，前者将骨骼长度映射到形状参数，后者将骨骼方向映射到姿态参数。此外，我们设计了两种自适应姿态细化机制，称为PR，以减轻误差累积问题。SMPL-IKS具有通用性，因此可扩展至MANO或SMPL-H。在针对仅身体、仅手部和身体-手部场景的各种基准测试中进行了广泛的实验。我们的模型在超越现有最先进方法的同时，效率也大大提高。数据和相关代码可在https://github.com/Z-Z-J/SMPL-IKS获取。