Yolo object detection speed on a Jetson TX2

Hi All,

I am running Yolo object detector using GPU on a Jetson TX2.

When I run the object detector for 200 images sequentially, I see often a sudden increase of detection speed.

All images are of same size and color channel.

The detection speed gets faster about 2 times and the starting point of speeding up is not consistent.

When it happens, the increased detection speed is maintained later.

I can’t figure out why this sudden speedup happens.

Any help would be appreciated!

Thanks.

Hi,

It sounds like you are in a dynamic mode.

Have you locked the device clock to max first?

sudo ./jetson_clocks.sh

Thanks.

Thanks.

It worked very well.

Now, the speed is 3 times faster continuously from the first image detection.

By looking at the “jetson_clocks.sh --show” command result,
the current frequency of 4 CPUs is now set to max, but 2 CPUs are set to about 90% of max frequency.

How can I set all the CPU frequency to max?

Would it be OK to use TX2 for more than 30 minutes setting the frequencies to max?

I figured out how to use “nvpmodel” and now can set 6 cores from 2 CPUs to maximum performance.

Thanks a lot.

Hi,

You can find more information about nvpmodel here:
https://www.jetsonhacks.com/2017/03/25/nvpmodel-nvidia-jetson-tx2-development-kit/

Thanks.