Reproduce “IMU-Bench: Benchmarking Wrist IMU Representations Across Kinematic, Semantic and Clinical Tasks” 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
Yiquan Li, Aidan Bradshaw, Chenfeng Gao, Karan Ahuja
IMU-Bench tests whether unlabeled wrist-motion representations transfer across activity, pose, odometry, gesture, clinical, and motion–text tasks. InrtlWrist reconstructs masked accelerometer and gyroscope measurements, then is evaluated with frozen probes and fine-tuning. It has best or joint-best frozen scores on nine of ten tasks; fine-tuning reaches 88.1 mm on Nymeria pose versus 92.7 mm from scratch, although participants recur across its session-held-out splits.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.