Wildlife fence road ecological dataset in South Africa

This dataset is a comprehensive dataset of wildlife fence images, containing both electric and non-electric fences, to support computer vision, machine learning, and deep learning research in classification, object detection, semantic segmentation, and instance segmentation. Images were collected using a drone and a still camera in natural sunlight. Images were captured at varying viewpoints, distances, and angles to add diversity and practicality, resized and saved in PNG format. For classification purposes, the dataset is divided into folders for single-fence and double-fence configurations and the images are stored in their respective folders. For segmentation tasks, images were zoomed in and resized to focus on fence parts such as insulators, and polygon-based annotations were generated using the VGG Image Annotator (VIA). These annotations were converted into binary masks to enable pixel-level learning tasks. Thus, the dataset consists of the original images, VIA JSON annotation files, and instance-level and semantic segmentation masks obtained during preprocessing. The dataset is accompanied by Python scripts that automate the conversion of VIA polygon annotations into binary masks and the organisation of the dataset into a machine-learning-ready directory structure. This enables researchers to easily adapt the dataset for applications like semantic segmentation, instance segmentation, object detection, object counting, and classification. This dataset can serve as a valuable basis for future research in road ecology, wildlife conservation, and intelligent transportation systems, especially for developing automated fence recognition and monitoring systems to reduce wildlife-vehicle collision hotspots.