WildlifeReID-10k: Wildlife re-identification dataset with 10k individual animals

• DOI: 10.48550/arxiv.2406.09211
• Publication Date: 2024-06-13
• Authors: Lukáš Adam; Vojtěch Čermák; Kostas Papafitsoros; Lukáš Picek
• Keywords: Identification; Wildlife trade

We introduce a new wildlife re-identification dataset WildlifeReID-10k with more than 214k images of 10k individual animals. It is collection 30 existing datasets additional processing steps. contains animals as diverse marine turtles, primates, birds, African herbivores, mammals and domestic Due to the ubiquity similar in datasets, we argue that standard (random) splits into training testing sets are inadequate for propose similarity-aware split based on similarity extracted features. To promote fair method comparison, include both closed-set open-set settings, use MegaDescriptor - foundational model baseline performance host leaderboard best results. publicly publish codes used create it wildlife-datasets library, making highly curated easy use.