Quantifying Motion Accuracy: MPJRE, MPJPE, and Consistency Metrics for HMD-based Avatar Generation

Table of Links

Abstract and 1. Introduction

2. Related Work

   2.1. Motion Reconstruction from Sparse Input

   2.2. Human Motion Generation

3. SAGE: Stratified Avatar Generation and 3.1. Problem Statement and Notation

   3.2. Disentangled Motion Representation

   3.3. Stratified Motion Diffusion

   3.4. Implementation Details

4. Experiments and Evaluation Metrics

   4.1. Dataset and Evaluation Metrics

   4.2. Quantitative and Qualitative Results

   4.3. Ablation Study

5. Conclusion and References

Supplementary Material

A. Extra Ablation Studies

B. Implementation Details

4. Experiments and Evaluation Metrics

4.1. Dataset and Evaluation Metrics

We train and evaluate our method on AMASS [25], which unifies multiple motion capture datasets [2, 4, 6, 12, 23, 26, 28, 37, 41–43] as SMPL [22] representations.

We report several metrics for evaluations and comparisons: mean per joint rotation error (MPJRE) and mean per joint position error (MPJPE) for measuring the average relative rotation and position error across all joints respectively, as well as the average position error of the root joints (Root PE), hand joints (Hand PE), upper-body joints (Upper PE), and lower-body joints (Lower PE).

Besides the above reconstruction accuracy, we also evaluate the spatial and temporal consistency of the generated sequences, as it significantly contributes to the visual quality. Specifically, we calculate the mean per joint velocity error (MPJVE) and Jitter, where MPJVE measures the average velocity error of all body joints, and Jitter quantifies the average jerk (time derivative of acceleration) of all body joints. In both cases, lower values indicate better results.