24 ms·
I noticed that you can use Tesseract as an OCR adapter for rga. Tesseract is written in python, IIRC, and in the OP it comes with a warning that it’s slow and n
by faitswulff 6y ago
I noticed that you can use Tesseract as an OCR adapter for rga. Tesseract is written in python, IIRC, and in the OP it comes with a warning that it’s slow and not enabled by default. Are there any other fast, reliable OCR libs out there? Or any rust OCR backends?
- mouldysammich 6y agohttps://github.com/tesseract-ocr/tesseract https://github.com/tesseract-ocr/tesseract seems to be written in c++ not python
- faitswulff 6y agoAh, my mistake then.
- hobofan 6y agoI don't think the problem necessarily is that Tesseract is slow, but that the whole process of rendering a PDF to a series of PNGs on which you can then run OCR is slow (which is what it does in the background).
- undebuggable 6y agoThe process of converting all pages to raster images and then OCR-ing each one takes hours for PDFs hundreds of pages long. This workflow is not suitable for instant search. For non OCR-ed PDFs it's worth to pregenerate the text.
- hobofan 6y agoThat's why rga comes with a cache. I've occasionally used the Tesseract adapter with good success (results-wise), and after the inital rendering and indexing, it's fast enough to use.