Tesseract OCR working slowly on Jetson Nano

[jeremy1403]

Hi everyone,

I’m actually working on an OCR which has to be deployed on a tablet outside. That’s why I bought a Jetson Nano since the last hardware wasn’t fast enough.

My project is using tesseract to do the OCR part. On my computer (Ryzen 5 3600 + RTX 2060) the entire pipeline is about 3 seconds long. But on the jetson nano it is more than 1 minute long which is really too long to be deployable.

So, I wanted to know if there were configurations or optimizations possible on the jetson nano to greatly reduce the time needed to perform the OCR.

Thank you,
Jeremy

[AastaLLL]

Hi, Jeremy

Have you maximized the performance first?

$ sudo nvpmodel -m 0
$ sudo jetson_clocks

Thanks.

[jeremy1403]

Hi, thanks for your answer,

I had not maximized the performance. It shortened by several seconds (maybe 20) but I need something around 5 seconds and 40 is still too long. Are there any others optimizations I need ? Especially for tesseract OCR (used via the pytesseract library in python).

Thank you

[dkreutz]

There is some experimental OpenCL support for tesseract-ocr: Tesseract and OpenCL | tessdoc

[jeremy1403]

We abandonned tesseract, you can close the topic, thanks for the answers !