Augmented and Diverse Herding Dataset for Autonomous Shepherd Robots

This dataset enables real-time object detection of sheep, wolves, dogs, wild dog, redfox, fox, coyote, cow and humans for robotic shepherding applications. Built from raw YOLOv5-style sources, it integrates class balancing, video-based diversity, and strong augmentations to enhance robustness. A recycling strategy is used for rare classes. Compatible with YOLOv5 to YOLOv12, RT-DETR, and ROS 2 deployments on legged robots, the dataset includes labels, images, statistics, and visualizations, ready for direct use in training detection models for autonomous livestock protection.