MorphCL: Morphological Contrastive Learning for Inertial-based Human Activity Recognition
Marius Bock, Yuwei Zhang, Juergen Gall, Michael Moeller, Kristof Van Laerhoven, Cecilia Mascolo (Univ. Bonn / Univ. Cambridge / Univ. Siegen / Lamarr Institute)
arXiv 2610.10245· 2026
IMU
Summary
A plug-in self-supervised loss for wrist-accelerometer pretraining with three steps:
Key facts
- Form Factor
- wrist_accelerometer
Reported benchmarks
Human activity recognition (IMU)
PAMAP2 · Macro-F1 ↑
#1 of 1
75.15 LP / 75.80 FT (%)
LOSO, right wrist 50 Hz 2.56 s, NULL kept; comparator Yuan et al. UKB ResNet18 (harnet-ukb) 77.06 LP / 84.00 FT on same protocol (paper Table 5).
Updated: Oct 8, 2026