SealID: Saimaa Ringed Seal Re-Identification Dataset

Area of Science:

• Wildlife biology and conservation
• Computer vision and machine learning

Background:

• Monitoring endangered species like the Saimaa ringed seal (Pusa hispida saimensis) is crucial for conservation efforts.
• Automated methods are needed to process large volumes of image data from camera traps and crowd-sourced material for population monitoring.
• Re-identification of individual animals using unique natural markings is a key task in population studies.

Purpose of the Study:

• To introduce and describe the Saimaa ringed seal image (SealID) dataset, comprising 57 images of individual seals.
• To propose an evaluation protocol for re-identification methods applied to Saimaa ringed seals.
• To provide baseline results for two existing re-identification methods (HotSpotter and NORPPA) on the SealID dataset.

Main Methods:

• Compilation of a novel image dataset (SealID) of Saimaa ringed seals.
• Development of an evaluation protocol for assessing animal re-identification algorithms.
• Application and evaluation of HotSpotter and NORPPA algorithms on the SealID dataset.

Main Results:

• The SealID dataset, containing 57 images of Saimaa ringed seals, is made publicly available.
• Baseline performance metrics for HotSpotter and NORPPA on the challenging Saimaa ringed seal re-identification task are reported.
• The dataset and proposed protocol serve as a benchmark for future development of re-identification methods for this endangered species.

Conclusions:

• The SealID dataset provides a valuable resource for advancing automated re-identification techniques for endangered freshwater seals.
• The study highlights the challenges in Saimaa ringed seal re-identification due to pose variation and pelage pattern appearance.
• Publicly releasing the dataset facilitates research into robust computer vision methods for wildlife monitoring and conservation.