🤖 AI Summary
To address the scarcity of labeled data for wildlife re-identification in non-urban environments, this paper proposes a temporal self-supervised learning framework tailored for camera-trap videos. The method leverages unlabeled consecutive video frames to model temporal consistency of individual appearance and employs contrastive learning to extract robust, view-invariant feature representations. Its key contribution is the first systematic integration of temporal self-supervision into open-world wildlife re-identification—enabling discriminative individual representation learning without manual annotations. Experiments demonstrate that the approach significantly outperforms supervised baselines across multiple species (e.g., leopard cats, wild boars), particularly excelling in few-shot and cross-domain generalization. Moreover, the learned features achieve state-of-the-art performance on downstream tasks including image retrieval, unsupervised clustering, and few-shot fine-tuning.
📝 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.