← ArchivePaper2026
SOMA: Unifying Parametric Human Body Models
Jun Saito, Jiefeng Li, Michael de Ruyter, Miguel Guerrero, Edy Lim, Ehsan Hassani, Roger Blanco Ribera, Hyejin Moon, Magdalena Dadela, Marco Di Lucca, Qiao Wang, Xueting Li, Jan Kautz, Simon Yuen, Umar Iqbal
SOMA introduces a canonical body topology and universal rig, reducing cross-model adapter complexity from O(M^2) to O(M) via inverse-LBS pose solving and FK optimization.
Abstract
Parametric human body models are foundational to human reconstruction, animation, and simulation, yet they remain mutually incompatible: SMPL, SMPL-X, MHR, Anny, and related models each diverge in mesh topology, skeletal structure, shape parameterization, and unit convention, making it impractical to exploit their complementary strengths within a single pipeline. We present SOMA, a unified body layer that bridges these heterogeneous representations through three abstraction layers. Mesh topology abstraction maps any source model's identity to a shared canonical mesh in constant time per vertex. Skeletal abstraction recovers a full set of identity-adapted joint transforms from any body shape, whether in rest pose or an arbitrary posed configuration, in a single closed-form pass, with no iterative optimization or per-model training. Pose abstraction inverts the skinning pipeline to recover unified skeleton rotations directly from posed vertices of any supported model, enabling heterogeneous motion datasets to be consumed without custom retargeting. Together, these layers reduce the $O(M^2)$ per-pair adapter problem to $O(M)$ single-backend connectors, letting practitioners freely mix identity sources and pose data at inference time. The entire pipeline is fully differentiable end-to-end and GPU-accelerated via NVIDIA-Warp.
How to read this
- Category
- Method: a unifying interoperability layer for parametric body models
- Contributions
- A canonical body topology with mesh topology abstraction that maps any source model's identity to a shared canonical mesh in constant time per vertex
- Skeletal abstraction recovering identity-adapted joint transforms from any body shape (rest or posed) in a single closed-form pass, with no iterative optimization or per-model training
- Pose abstraction that inverts skinning (inverse-LBS) to recover unified skeleton rotations from posed vertices, collapsing the O(M^2) per-pair adapter problem to O(M) single-backend connectors
- Context
- It sits on top of and bridges established parametric body models (SMPL, SMPL-X, the Momentum Human Rig (MHR), and Anny) and motion archives like AMASS, aiming to make their complementary strengths usable in one pipeline.Builds on: SMPL: A Skinned Multi-Person Linear Model · Expressive Body Capture: 3D Hands, Face, and Body from a Single Image · AMASS: Archive of Motion Capture as Surface Shapes · MHR: Momentum Human Rig · Human Mesh Modeling for Anny Body
- Correctness
- The approach assumes the heterogeneous source models (differing in topology, skeleton, shape parameterization, and units) can be reconciled through a shared canonical representation and an invertible skinning pipeline; readers should note it is demonstrated on the specific supported models listed, and fidelity of the inverse-LBS pose recovery and closed-form joint solve under extreme shapes or out-of-distribution poses is the natural thing to scrutinize.
- Clarity
- Reasonably accessible at a high level (the three-layer abstraction framing reads cleanly); a first pass conveys the unifying idea, but a second pass is needed to follow the closed-form skeletal solve and the inverse-skinning formulation.
- How to read it
- First pass: read the abstract and fix the three abstraction layers (mesh, skeletal, pose) and the O(M^2) to O(M) framing in mind. Do a second pass on the inverse-LBS pose solving and the closed-form FK/joint-transform recovery if you intend to implement retargeting or mix motion datasets; check which models are actually supported before assuming coverage.
Builds on
Built upon by
Nothing yet.
This page summarises the entry and links to its original source. The archive never hosts or redistributes the publication itself.Show it in the full archive list →