An unoptimized PyTorch model may achieve only 2–3 FPS on the Jetson Nano, which can be improved to 10–25 FPS after TensorRT optimization, an improvement of up to 5–8x