Reproduce “NymeriaPlus: Enriching Nymeria Dataset with Additional Annotations and Data” 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
Daniel DeTone, Federica Bogo, Eric-Tuan Le, Duncan Frost, Julian Straub, Yawar Siddiqui, Yuting Ye, Jakob Engel and 2 more
Meta Reality Labs Research introduces NymeriaPlus, an extended version of the Nymeria dataset, providing refined full-body human motion in MHR and SMPL formats, dense 3D and 2D object annotations, instance-level 3D object reconstructions, and additional multimodal data streams. The enhancements reduce human motion artifacts, with wrist distance error decreasing to 5.07 cm and body self-penetration error to 2.44, and provide 4,547 high-quality 3D object shapes.
Key insights
- A joint optimization approach combining inertial motion capture (XSens) with egocentric wearable data (Aria headset and wristbands) effectively reduces common motion artifacts like foot sliding and self-penetrations, improving human pose accuracy.
- Leveraging static 'basemap' recordings for initial dense 3D object annotation and then propagating and refining these annotations across dynamic egocentric sequences is an efficient method to enrich scene context at scale.
- Integrating state-of-the-art 3D reconstruction techniques (ShapeR) with a multi-stage manual quality control process enables the generation of high-fidelity, instance-level 3D object geometries from egocentric imagery.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.