Wildlife Target Re-Identification Using Self-supervised Learning in Non-Urban Settings

TL;DR

This paper explores self-supervised learning for wildlife re-identification, demonstrating that it outperforms supervised methods in robustness and effectiveness across various tasks without requiring labeled data.

Contribution

It introduces a self-supervised approach using temporal image pairs for wildlife re-identification, reducing dependence on annotated datasets and improving performance.

Findings

Self-supervised models are more robust with limited data.

Self-supervised features outperform supervised features in downstream tasks.

The method leverages camera trap video data without labels.

Abstract

Wildlife re-identification aims to match individuals of the same species across different observations. Current state-of-the-art (SOTA) models rely on class labels to train supervised models for individual classification. This dependence on annotated data has driven the curation of numerous large-scale wildlife datasets. This study investigates self-supervised learning Self-Supervised Learning (SSL) for wildlife re-identification. We automatically extract two distinct views of an individual using temporal image pairs from camera trap data without supervision. The image pairs train a self-supervised model from a potentially endless stream of video data. We evaluate the learnt representations against supervised features on open-world scenarios and transfer learning in various wildlife downstream tasks. The analysis of the experimental results shows that self-supervised models are more robust even with limited data. Moreover, self-supervised features outperform supervision across all downstream tasks. The code is available here https://github.com/pxpana/SSLWildlife.