Leveraging Anthropometric Measurements to Improve Human Mesh Estimation and Ensure Consistent Body Shapes

TL;DR

The paper addresses frame-to-frame inconsistencies in the basic body shape produced by SOTA human mesh estimation by introducing A2B, a learned mapping from 36 anthropometric measurements to SMPL-X shape parameters. By measuring a person once and applying the resulting A2B shape parameters across frames, the approach yields consistent meshes and improves pose accuracy when combined with inverse kinematics and (optionally) sequence-based 3D HPE uplifts. It demonstrates substantial improvements on sports datasets ASPset and fit3D, including MPJPE reductions exceeding several tens of millimeters over traditional HME baselines, and shows that incorporating anthropometric guidance also benefits existing HME models. The work highlights GT inconsistencies in current datasets and provides a practical, measurement-driven pathway to achieve accurate, stable 3D human meshes suitable for sports analysis and performance assessments.

Abstract

The basic body shape (i.e., the body shape in T-pose) of a person does not change within a single video. However, most SOTA human mesh estimation (HME) models output a slightly different, thus inconsistent basic body shape for each video frame. Furthermore, we find that SOTA 3D human pose estimation (HPE) models outperform HME models regarding the precision of the estimated 3D keypoint positions. We solve the problem of inconsistent body shapes by leveraging anthropometric measurements like taken by tailors from humans. We create a model called A2B that converts given anthropometric measurements to basic body shape parameters of human mesh models. We obtain superior and consistent human meshes by combining the A2B model results with the keypoints of 3D HPE models using inverse kinematics. We evaluate our approach on challenging datasets like ASPset or fit3D, where we can lower the MPJPE by over 30 mm compared to SOTA HME models. Further, replacing estimates of the body shape parameters from existing HME models with A2B results not only increases the performance of these HME models, but also guarantees consistent body shapes.

Figures (11)

• Figure 1: Two qualitative examples from the ASPset sports dataset. The result from a SOTA HME model, SMPLer-X smplerx, is shown on the left, the result from our model on the right, respectively. GT joints and estimated joints are color-coded. Corresponding joints are connected.
• Figure 2: Histograms and fitted normal distribution (orange) for the first two $\beta$ parameters for all male (left) and female (right) subjects of the AGORA agora dataset.
• Figure 3: Overview of our inference pipeline. The pose and shape parameters are obtained either from IK applied to UU results (Sec. \ref{['sec:ik_res']}) or from an HME model (Sec. \ref{['sec:hme_res']}). In real applications, the anthropometric measurements will be taken directly from the humans. For our evaluations, we use the GT shape parameters and further experiment with the shape parameters of the respective model (IK or HME).
• Figure 4: Visualization of the used landmarks with a standard T-pose SMPL-X mesh in front view.
• Figure 5: Visualization of a subset of the used landmarks in side view.