ATLAS: Decoupling Skeletal and Shape Parameters for Expressive Parametric Human Modeling
Jinhyung Park, Javier Romero, Shunsuke Saito, Fabian Prada, Takaaki Shiratori, Yichen Xu, Federica Bogo, Shoou-I Yu, Kris Kitani, Rawal Khirodkar
2025-08-22
Summary
This paper introduces a new, more detailed 3D model of the human body called ATLAS, designed to realistically represent a wide variety of body shapes and poses.
What's the problem?
Current 3D body models aren't great at capturing the full range of human body diversity, meaning they struggle with accurately representing people in different poses or with different body types. This is often because they don't have enough diverse data to learn from, and they make assumptions that limit how much detail they can show. Also, many models try to build the body's surface first and *then* figure out where the skeleton should be, which creates a weird connection where changes to the skin affect the bones, and vice versa, making it hard to control things like height or limb length.
What's the solution?
The researchers created ATLAS by using a huge dataset of 600,000 detailed 3D scans of people taken with many cameras at once. The key is that they separated how the body *shape* is modeled from how the *skeleton* is modeled. Instead of building the skin and then the bones, they started with the skeleton and built the body around it. This allows for more realistic shapes, easier customization of body features, and more accurate positioning of joints, regardless of how much muscle or fat is on the outside.
Why it matters?
ATLAS is a significant improvement because it can more accurately fit to new people in different poses than previous models. It’s better at handling complex poses and allows for more control over body characteristics, which is important for applications like animation, virtual reality, and even medical simulations where realistic human representation is crucial.
Abstract
Parametric body models offer expressive 3D representation of humans across a wide range of poses, shapes, and facial expressions, typically derived by learning a basis over registered 3D meshes. However, existing human mesh modeling approaches struggle to capture detailed variations across diverse body poses and shapes, largely due to limited training data diversity and restrictive modeling assumptions. Moreover, the common paradigm first optimizes the external body surface using a linear basis, then regresses internal skeletal joints from surface vertices. This approach introduces problematic dependencies between internal skeleton and outer soft tissue, limiting direct control over body height and bone lengths. To address these issues, we present ATLAS, a high-fidelity body model learned from 600k high-resolution scans captured using 240 synchronized cameras. Unlike previous methods, we explicitly decouple the shape and skeleton bases by grounding our mesh representation in the human skeleton. This decoupling enables enhanced shape expressivity, fine-grained customization of body attributes, and keypoint fitting independent of external soft-tissue characteristics. ATLAS outperforms existing methods by fitting unseen subjects in diverse poses more accurately, and quantitative evaluations show that our non-linear pose correctives more effectively capture complex poses compared to linear models.