JMM-TGT: Self-supervised 3D action recognition through joint motion masking and topology-guided transformer
Han Wen, Guangping Zeng, Qingchuan Zhang, Zihan Li, Mengyang Zhu · 2025
Read JMM-TGT: Self-supervised 3D action recognition through joint motion masking and topology-guided transformer by Han Wen, Guangping Zeng, Qingchuan Zhang, Zihan Li, Mengyang Zhu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-12752959.txt.
Description
Han Wen and co-authors taught machines to read skeleton actions without labels in two thousand twenty-five. Action recognition from joint tracks has many uses. Labels for training are slow to produce. Their model learns by hiding joints and guessing them. Which joints hide depends on motion patterns. Similar and different movements set the hiding odds.
A transformer guided by body topology does the learning. Joint connections shape its attention structure. Subtle motion differences become easier to catch. The model builds rich features before any labels arrive. Fine-tuning then needs far fewer examples. Surveillance and rehabilitation tools both use such models. Self-supervision cuts their data costs sharply.
Catalog record
| CREATOR | Han Wen, Guangping Zeng, Qingchuan Zhang, Zihan Li, Mengyang Zhu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 55 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |