Comparative Evaluation of YOLO-family Detectors for Pig Detection in Precision Livestock Systems

Precision Livestock Farming (PLF) delivers technological solutions for modern pork production, where real-time monitoring is vital for animal health and welfare. Object detection models are key in PLF, yet comprehensive baselines for recent architectures on specialized datasets remain limited. This study benchmarks two state-of-the-art models, YOLOv8 and YOLOv12, across five sizes (nano to extra-large) for pig detection using the public PigLife dataset. Ten models were fine-tuned and evaluated with standard COCO metrics. Results show that while accuracy improves with model size, gains diminish with larger models. The YOLOv8m, with 25.9 million parameters, achieved a near-peak mean Average Precision (mAP) of 91.8%. This work establishes a new benchmark on PigLife, surpassing previous results, and highlights the trade-off between YOLOv8’s higher accuracy and YOLOv12’s parameter efficiency.