How to implement a License Plate Detection & Recognition in DeepStream Python?

I have an LPR system which I implemented in python pytorch.
And I found that DeepStream is such a powerful system for processing stream and AI application,

Also DeepStream has a very clear code written with a pipeline.
So I think there should be a way to replace the detection or recognition part of the pipline.

But I have study the DeepStream python version bellow for a while.

I have no idea how to implement my own LPR programe into this DeepStream python.

Including Image pre-processing, Dection&Recognition proecesses

I want some hints or advise!!
Thank you so much!!

  1. please share your setup with us
  2. You can refer GitHub - NVIDIA-AI-IOT/deepstream_lpr_app: Sample app code for LPR deployment on DeepStream
  1. My setup:
    OS == Ubuntu18.04
    python == 3.6.8
    torch == 1.7.1
    tensorrt ==
    Cuda == 10.2

  2. My application structure might be similar with the picture bellow.
    But the “Car detection”,“License Plate detection” and “Plate recognition” parts are all written in Python Pytorch in my own program.
    So is there any possible to implement my program into DeepStream-Python-API?

Yeah, I think it’s possible to do that.