Jetson Shutdown while running SSDLite Model with multithreading


I am using MobileNetSSDLite Model to detect custom object, it worked pretty well on my i3 PC but in jetson it takes around 5 min to load the lite model and after that i have multithreaded the rtsp video stream to get the frames but as soon as it starts the first frame appears and it crashes.

Considering two factor that might affect the shutdown are:
1-Power starvation - for this connected barrel jack and provided 5v 5amp SMPS
2-RAM constrain - added a Swap file of 4GB

Still it shuts down, If i remove the multithreaded for getting the frame the feed lag a lot and after some time it also crashes

Using Opencv 4.0(build from source) and tensorflow 1.13


Would you mind to set the Nano into 5W mode and monitor the memory with tegrastats?

sudo nvpmodel -m 1
sudo tegratstats

Suppose you can find the exact memory required from the model with tegrastats info.

By the way, please noticed that the TensorFlow is not an optimized inference solution on Jetson.
It’s recommended to convert the model into .uff and execute it with TensorFlow instead.