Speed up

Hello everyone.
I am using opencv (4.5.4) to inference on a pre-trained model. In order to use the GPU of my jetson nano, the following commands are present in my code:

net.setPreferableTarget (cv2.dnn.DNN_TARGET_CUDA)

In addition to these two commands, I also use:
sudo nvpmodel -m 0
sudo jetson_clocks

From a performance point of view, I can’t achieve the same performance achieved by SegNet inference (in terms of fps).
I think TensorRT is not being taken advantage of by the commands I just reported, but I could be wrong.
Do you have any suggestions for speeding up inference in opencv?

1 Like


TensorRT is a stand-alone library that different from the CUDA implementation.
To run a model with TensorRT, please use TensorRT API directly.


This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.