Reproduce “IMUDINO: Self-distilled Foundation Model for Inertial Sensing” 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
Zhoutong Ye, Chengwen Zhang, Zixuan Wang, Xufeng Jian, Zhe He, Qingyang Wan, Xutong Wang, Jiaqi Liu and 4 more
IMUDINO adapts DINO self-distillation and iBOT++ to learn global and time-local representations from varied inertial data, using motion transformations and sampling-rate-aware positional encoding. Its frozen 85M backbone, pretrained on wearable data, transfers to robot odometry: a probe trained on labeled robot data achieves 0.209 m/s validation average velocity error, below TartanIMU experts’ 0.350 m/s; this is not zero-shot.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.