I have been working on deepstream but the existing LPR models are trained or US dataset or CCPD so , they are giving gibberish results on the indian numberplates but perform very well on the US styled number plates , there is also no such open source datasets that may facilitate to train a model of my own , but realistically training such model would be very demanding and computationally expensive. Furthermore , the setting up of different ocr models like monkt/paddleocr-onnx( monkt/paddleocr-onnx · Hugging Face ) is also not very simple as it requires either heavy surgical implementations on the SDK or a python script which runs the ocr model from extracted number plates from the lpd. I am still very new to the SDK and am not much experienced. I hope that there may exist some solutions for a problem like this. For more context I am working on Tesla T4 based server.
Hardware Platform :- Tesla T4
DeepStream version :- 7.0
Cuda version :- 12.4
Driver Version :- 550.127.05