Reproduce “Improving 3D Foot Motion Reconstruction in Markerless Monocular Human Motion Capture” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tom Wehrbein, Bodo Rosenhahn
FootMR refines ankle motion in markerless 3D human-motion capture to improve foot poses in reconstructions. It uses detected 2D foot points and initial ankle and knee rotations to predict residual ankle corrections while retaining an existing full-body estimator. On MOYO, global ankle joint angle error falls from 29.6° to 25.9° for GVHMR23j with FootMR; the method refines ankles, not toe articulation.
Key insights
- Decoupling 3D foot motion refinement from the main body motion estimation allows for more focused and accurate capture of fine-grained foot articulations.
- Predicting residual global ankle rotations, informed by initial global ankle and knee rotations, effectively resolves the inherent ambiguities in 2D-to-3D lifting and enhances robustness to noisy 2D keypoints.
- Utilizing synthetic 2D keypoints derived from precise 3D motion capture data for training bypasses the limitations of inaccurate 3D foot annotations in existing video datasets, improving model performance on foot-specific movements.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.