Identify the sources of the DanceHMR performance discrepancy

Identify which changes to body features, training data, keypoint detection, implementation, or evaluation account for the discrepancy between the published DanceHMR results and the results obtained with the locally implemented DanceHMR-style fusion baseline.

Background

The paper reports that its locally implemented DanceHMR-style fusion baseline achieves lower ARCTIC errors than the published DanceHMR results. Because the body features, training mixture, and keypoint detector were changed jointly, and because some implementation and evaluation details of the published method are unspecified, the paper cannot determine which factors cause the observed discrepancy. Isolating these sources would improve the interpretability and reproducibility of the comparison.

References

Because the body features, training mixture, and keypoint detector were changed jointly, and the publication leaves some implementation and evaluation details unspecified, we cannot isolate the sources of this discrepancy.

— ExMR: Extremity Motion Refinement for Monocular Human Motion Capture  (2610.26213 - Wehrbein et al., 19 Oct 2026) in Section 4.1, subsection “Expressive Whole-Body and Hand Reconstruction”