Why do I test mobienet-v2 with Jetson nano,The actual FPS are quite different from the official ones?

Why do I test mobienet-v2 with Jetson nano,The actual FPS are quite different from the official ones.

the CUDA operation time is not included?just consider only CPU operation time

the official fps is 27,but the real fps is 5~7

Hi,

Our official data is the inference time.
Please noticed that we don’t run mobienet-v2 with TensorFlow but accelerated with TensorRT.
You can check this topic for the instructions of bench-marking first:
https://devtalk.nvidia.com/default/topic/1050377/jetson-nano/deep-learning-inference-benchmarking-instructions/

Thanks.