Abstract
1. Photo-identification of individual snow leopards (Panthera uncia) is the primary technique for density estimation for the species. A high volume of images from multiple projects, combined with pre-existing historical catalogs, has made identifying snow leopard individuals within the images cost- and time-intensive. 2. To speed the classification among a high volume of photographs, we trained and evaluated image classification methods for PIE v2 (a triplet loss network), and we compared PIE’s accuracy to the HotSpotter algorithm (a SIFT based algorithm). Analyzed data were collected from a curated catalog of free-ranging snow leopards photographed across years (2012-2019) in Afghanistan and from samples in captivity provided by zoos from Finland, Sweden, Germany, and the United States. 3. Results show that PIE outperforms HotSpotter. We also found weaknesses in the initial PIE model, like a minor amount of background matching, which was addressed, although likely not fully resolved, by applying background subtraction (BGS) and left-right mirroring (LR) methods. The PIE BGS LR model combined with Hotspotter showed a Rank-1: 85%, Rank-5: 95%, Rank-20: 99%. 4. Overall, our results recommend implementing PIE v2 simultaneously with HotSpotter on Whiskerbook.org.
Competing Interest Statement
The authors have declared no competing interest.
