Very cool, I'm building my own local-first product as well

thank you! what is it about?

Curious how it does on multi-page scanned PDFs vs. single screenshots? The ORT vision/decoder split is the part that usually makes or breaks CPU VLM OCR...

I had to extract the image from a PDF for it to work. Then run it on each page image extracted.

Thanks

Now this is legit cool, keep up the great work.

thank you!

- how well do you think this ll work with code? i mean take code screenshots and convert it into actual code for vscode

Just ran

  textsnap "https://i.ytimg.com/vi/LBNDfxjEYlA/maxresdefault.jpg"

and got this

  $('.count').each(function () {
  $('this').prop('Counter', 0).animate({
    Counter: $('this').text()
  }, {
      duration: 4000,
      easing: 'swing',
      step: 'function (now) {
          $('this").text(Math.ceil(now));
      }
    }); 
  });

Roman alphabet only or does this work with other alphabets?

109 languages, including other alphabets.

What was the reason for adopting PaddleOCR? Can other OCR models be used as well?

No reason other than their Q4 model working reasonably well and fast on my CPU laptop. Should work with any ONNX VLM model

What's the performance like compared to tesseract? I don't see tesseract mentioned anywhere in the readme, which is surprising considering that's the number one tool most go to for Image > text OCR.

No rigorous eval, and I love Tesseract. Here's the example that motivated me to build textsnap (which is in the github's README), parsed with Tesseract:

https://imgur.com/a/i2eQra8

Very noticable difference and the exact issue I run repeatedly with tesseract! Definitely going to try dropping textsnap into my scripts now. Thanks!!

This is awesome! Been needing something like this for some research paper diagrams I've been indexing.