The model killed when I try to run torch2trt converted file

We succeeded in converting to torch2trt with the pth file of the slim model by following

It comes out as below when tested.

There seems to be no problem with the conversion, but if I put the above file into the entire file I want to run, the problem occurs. (The slim model is a model that finds landmarks on the face. After finding a face through object detection, it finds a landmark.) However, when this model is called in and executed, it takes a very long time and eventually killed.

I want to shorten the execution time by importing the converted file, what should I do?

As a result of checking with jtop, turning only the converted slim model takes up ram 1.9 GB. Since the Jetson Nano I am currently using is 2GB, it seems that doing object detection and face detection does not work and is killed. In this case, is there a solution to run two models?


We recommend you to please post your concern on Issues · NVIDIA-AI-IOT/torch2trt · GitHub to get better help.

Thank you.