Jetson nano Not able to give higher FPS

Autonomous Machines Jetson & Embedded Systems Jetson Nano
1

Hi,

We have a code for object tracking using YOLOv4-Tiny that we are running on Jetson Nano 2gb as well as Jetson Nano 4gb. We upgraded to 4gb as the FPS on 2gb was 1.7-1.8.
After running the codes on 4gb Jetson Nano we are still getting the same FPS as that of 2gb Jetson Nano.
How can I Increase FPS??

3

Hi,

First, have you maximized the device performance?

$ sudo nvpmodel -m 0
$ sudo jetson_clocks

Then you can check the below sample from our user.
He can reach 25.5 (416x416) or 36.6 (288x288) fps with YOLOv4-Tiny on Nano.

Thanks.

4

I tried to put down commands but no change in output and with respect to example code from github I am not able to find specific reason why am I getting 1-2 fps

5

Hi,

Which source do you use?
If you are using a camera input, would you mind switching to video data to see if any improvement?

Thanks.

6

I have tried both ways using videos and camera inputs and still not able to get enough speed

7

Hi,

Support you are using TensorRT with FP16 mode, is that correct?
If yes, could you monitor the device with tegrastats to see if the GPU resources are fully utilized?

Thanks.

8

I am using similar codes used in this repo…What will be the reason cause I am not using TensorRT with FP16 mode??
GPU is not used by the system after I check status GPU was at 0% use

9

Hi,

There are several frameworks that can support DNN inference.
Based on the GitHub source, it deploy it with TensorFlow rather than TensorRT.

For Jetson, it’s recommended to convert the model into TensorRT for acceleration.
Below is an example of YOLOv4 + TensorRT for your reference:

Thanks.