WHIP

Towards Real-World Wearable Motion Reconstruction

ECCV 2026

1Max Planck Institute for Informatics, Saarland Informatics Campus, Germany

2Saarbrucken Research Center for Visual Computing, Interaction and AI, Germany

3Google, Switzerland

WHIP wearable capture setup and reconstructed full-body motion

TLDR: WHIP studies real-world wearable motion capture from arbitrary combinations of phones, watches, insoles, and headset tracking.

Abstract

The modern-day surge in popularity of wearable devices poses a fundamentally unique motion capture problem: reconstructing full-body movement from any set of sensing hardware worn at a given moment. Most research efforts assume fixed sensor configurations, such as IMU suits or HMD-centric rigs, and cannot generalize across them. WHIP prioritizes unobtrusive, lightweight devices such as smartphones, smartwatches, smart glasses, and smart insoles, and studies how their signals complement one another.

We introduce a large-scale multimodal dataset that synchronizes consumer-grade wearables with ground-truth 3D motion. The dataset spans 14 participants, two sessions per participant, 50 action classes, and more than 7 hours of synchronized recordings. We also propose WHIP, a flow-matching generative model that reconstructs physically plausible full-body motion from arbitrary subsets of available sensors.

Examples

Model predictions across different sensor conditioning.

Ground Truth

Reference motion

Prediction #1

Label Insoles L Watch L Phone

Prediction #2

Label Insoles L Watch L Phone + R Watch

BibTeX Citation

    @InProceedings{Camiletto_2026_WHIP,
  author    = {Boscolo Camiletto, Andrea and Dabral, Rishabh and Alvarado, Eduardo and Beeler, Thabo and Habermann, Marc and Theobalt, Christian},
  title     = {Towards Real-World Wearable Motion Reconstruction},
  booktitle = {Proceedings of the European Conference on Computer Vision (ECCV)},
  year      = {2026}
}