Wildlife Target Re-Identification Using Self-supervised Learning in Non-Urban Settings
Mufhumudzi Muthivhi, Terence L. van Zyl · 2025
Read Wildlife Target Re-Identification Using Self-supervised Learning in Non-Urban Settings by Mufhumudzi Muthivhi, Terence L. van Zyl (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/arxiv-2507-02403.txt.
Description
Mufhumudzi Muthivhi and Terence van Zyl test whether animals in camera-trap photos can be re-identified without any labeled examples. Naming individuals across sightings is central to wildlife monitoring, but labeling enough photos to train supervised models is slow and costly. Their approach pulls pairs of frames of the same animal from video over time, giving the model two views to learn from with no human annotation.
A self-supervised model trained on those pairs learned features that they then tested on open-world matching and on transfer to other wildlife tasks. The self-taught features proved more robust when data was scarce, and they beat supervised features on every downstream task tried. Because camera traps produce an endless stream of video, the method can keep learning from unlabeled footage. The authors release their code, aiming at field programs that cannot afford large labeling efforts.
Catalog record
| CREATOR | Mufhumudzi Muthivhi, Terence L. van Zyl |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 30 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-02403