search

NVIDIA-AI-IOT / Formula1Epoch

An autonomous R.C. racecar which detects people.
66 stars 22 forks source link

Performance considerations - using TensorRT? #3

Open mindThomas opened 6 years ago

mindThomas commented 6 years ago

It seems like you are not using TensorRT for inference? Wouldn't you be able to get better performance by using TensorRT rather than Keras for inference? What is the current performance / FPS achievable with your network running on the Jetson?

aayalur-nv commented 6 years ago

We were able to get around 4-5 hz for inference, sometimes a bit less. TensorRT would ideally make it a lot better, but for our purposes we thought that keras would make our concat network more modular and easier for us and other viewers to understand.