Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2602.11064 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912897966276608 |
|---|---|
| author | Darwish, Firas Nicholson, George Doherty, Aiden Yuan, Hang |
| author_facet | Darwish, Firas Nicholson, George Doherty, Aiden Yuan, Hang |
| contents | Synthetic data offers a compelling path to scalable pretraining when real-world data is scarce, but models pretrained on synthetic data often fail to transfer reliably to deployment settings. We study this problem in full-body human motion, where large-scale data collection is infeasible but essential for wearable-based Human Activity Recognition (HAR), and where synthetic motion can be generated from motion-capture-derived representations. We pretrain motion time-series models using such synthetic data and evaluate their transfer across diverse downstream HAR tasks. Our results show that synthetic pretraining improves generalisation when mixed with real data or scaled sufficiently. We also demonstrate that large-scale motion-capture pretraining yields only marginal gains due to domain mismatch with wearable signals, clarifying key sim-to-real challenges and the limits and opportunities of synthetic motion data for transferable HAR representations. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_11064 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Motion Capture is Not the Target Domain: Scaling Synthetic Data for Learning Motion Representations Darwish, Firas Nicholson, George Doherty, Aiden Yuan, Hang Machine Learning Synthetic data offers a compelling path to scalable pretraining when real-world data is scarce, but models pretrained on synthetic data often fail to transfer reliably to deployment settings. We study this problem in full-body human motion, where large-scale data collection is infeasible but essential for wearable-based Human Activity Recognition (HAR), and where synthetic motion can be generated from motion-capture-derived representations. We pretrain motion time-series models using such synthetic data and evaluate their transfer across diverse downstream HAR tasks. Our results show that synthetic pretraining improves generalisation when mixed with real data or scaled sufficiently. We also demonstrate that large-scale motion-capture pretraining yields only marginal gains due to domain mismatch with wearable signals, clarifying key sim-to-real challenges and the limits and opportunities of synthetic motion data for transferable HAR representations. |
| title | Motion Capture is Not the Target Domain: Scaling Synthetic Data for Learning Motion Representations |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2602.11064 |