Optimizing preprocessing and postprocessing of the Yolov5 neural network using CUDA, Thrust and Nvidia Performance primitives