How to get Jetson TX2 high fps? with YOLOV4

Robotics & Edge Computing Jetson Systems Jetson TX2
,
1

Hi, I have Jetson TX2 device.
When I run the model I trained with Yolov4-Tiny on my own computer, I get 28fps.
Graphics card model of my computer: GTX950M
When I run it with Jetson TX2 I get 10fps.
I’m using CUDA+CUDNN+OPENCVDNN on both devices.

I have a few questions regarding this.

  1. Is there a way or documentation to get higher fps for Jetson TX2 (yolov4 or yolov4-tiny)?
  2. How come I get higher fps on my own computer? Shouldn’t the Jetson TX2 give a higher result? Can anyone explain this?
  3. Is there a usage instruction for Jetson TX2?
3

Hi,

Have you maximized the device performance?

$ sudo nvpmodel -m 0
$ sudo jetson_clocks

On Jetson, it’s more recommended to use Deesptream as an inference pipeline.

This is because OpenCV might use CPU for processing, which is not an optimal way on Jeston.
But Deepstream will utilize some encoder hardware and GPU that can give you a better performance.

You can find an example for YOLOv4 with Deepstream below:
(tiny version should be very similar)

Thanks.