How to increase the speed of processing?

[VMM-MMV]

I trying to compile the models, as per this .

But I see no real difference in processing speed.

Are there some other ways to improve the speed of the models?

I don't particularly care about high accuracy, I just need the processing of pdf to txt to be fast.

[felixdittrich92]

Hi @VMM-MMV 👋,

Then you should try: https://github.com/felixdittrich92/OnnxTR :)

[felixdittrich92]

latency benchmark:

https://github.com/felixdittrich92/OnnxTR#benchmarks

I would suggest to try the standard installation onnxtr[cpu] first instead of onnxtr[openvino] because what i saw openvino differs heavily from CPU to CPU

with the following:

from onnxtr.models import ocr_predictor

model = ocr_predictor(
    det_arch='db_mobilenet_v3_large',  # detection architecture
    reco_arch='crnn_mobilenet_v3_small',  # recognition architecture
    det_bs=2, # detection batch size adjust to your hardware
    reco_bs=512, # recognition batch size adjust to your hardware
    )

You will get the fastest combination (maybe not the highest acc) ^^

[VMM-MMV]

Thank You. But, I am have some problems with it. Should I make a new discussion or should I ask here?

[felixdittrich92]

Opening a new would be good 👍