trying to do yolo on Jetson nano,but resulting nano shut down everytime due to memory problem.

Autonomous Machines Jetson & Embedded Systems Jetson Nano
#1

trying to do yolo on Jetson nano,but resulting nano shut down everytime due to memory problem.

I did my job following the link

it’s using tensorflow and I get out of memory failure with nano shut down automatically.

my question is:
how to use the tensort with it?
or should I just use the exe version make the train and inference?

by the way,there is no nano info in
https://github.com/NVIDIA-AI-IOT/deepstream_reference_apps/blob/master/yolo/README.md
does that mean it’s not recommended?

#2

The Nano doesn’t have a lot of RAM, and also has a very weak GPU compared to desktop GPUs and higher-end devices like the Xavier, so you will probably not be successful doing training on the Nano.
You should do your training on Windows, if you have “exe” versions of those bits.

Once you have a trained network, you may perhaps be able to load that model and run inference on the Nano – the only way to find out whether that will work is to try it!

#3

Hi,

Running out of memory won’t cause device shutdown, it should be a segmentation fault error.
Shut down is more likely caused by the power starvation.

What kind of power supply do you use?
Here is a power supply recommendation for your reference:
https://devtalk.nvidia.com/default/topic/1048640/jetson-nano/power-supply-considerations-for-jetson-nano-developer-kit/

Or you can try to use 5W mode to see if helps.

sudo nvpmodel -m 1

Thanks.

#4

thanks and I mean that, someone told me that he tried the original yolo v3 inference model on jetson nano at 5fpg, so I tried but with no success. It said “ran out of memory” like the following pic.
https://github.com/vxgu86/configSet/blob/master/imgs/nano-memory.jpg.
so I am trying yolo with tensorRT , my problem is if I run the original darknet model without tensorrt mentioned, would it be accelerated?

Thanks guys.

honestly speaking I ran into another problem with Jetson nano&& Jetpack 4.2.
it’s here, could you guys help solving it?

https://devtalk.nvidia.com/default/topic/1056384/jetson-nano/flushing-the-jetson-nano-amp-128g-sd-card-resulting-in-14-7g-available-and-shut-down-automatically-when-running-model/

#5

the power supply is 5V 3A, any problem?

#6

Hi,

Have you tried to setup Nano into 5W mode?
Usually, we are using the 5V⎓4A power supply in 10W mode.

Thanks.