I test the Jetson Xavier NX devkit performance. Use yolov4 model, precision is FP16, resolution is 416x416, and the inference result is 104ms per frame. Is is normal? Maybe too low? How to improve that?
Do you use TensorRT for inference?
More, please check if you have maximized the device with the following command:
$ sudo nvpmodel -m 0 $ sudo jetson_clocks
Already TensorRT for inference. GPU frequency is normal.
BTW, I can see many yolov3 benchmark result in github, but not yolov4. How to verify the yolov4 benchmark in Jetson Xavier NX. Are there some official suggestions?
We don’t have the benchmark data for YOLOv4.
But below is a tutorial to deploy it with TensorRT or TensorRT+Deepstream:
In general, we expect a similar performance between YOLOv4 and YOLOv3.
For XavierNX, YOLOv3 can achieve more than 30fps.