Comparative Analysis of YOLOv5 and YOLOv6 Models Performance for Object Classification on Open Infrastructure: Insights and Recommendations

In this paper, the performance of YOLOv5 and YOLOv6 models is evaluated on standardized object classification tasks using Kaggle open-access cloud processing infrastructure. The performance of these architectures is evaluated on a balanced subset of COCO dataset considering precision, recall, mAP0.5, and mAP0.5:0.95 metrics, learning and processing time. Our results indicate that the choice of the optimal model in the rapidly evolving field of computer vision and machine learning is heavily influenced by the dataset size, complexity, and restrictions imposed by the data processing platform. It is not always advisable to select the most recent algorithm for all applications. The obtained results, program code and the implemented architectures are available for non-commercial use at: https://github.com/mhorvat/yolov5_yolov6_comparison