Need Google Colab Tutorial for this

[ghost]

Hello, I hope you are doing well. I want a google colab tutorial for this repo

[felixdittrich92]

Hi @hammadyounas2008 👋,
we have some small Colab tutorials: https://mindee.github.io/doctr/notebooks.html

[ghost]

The saved pdf is not annotated? How can i download annotated pdf?

[felixdittrich92]

Mh PDF/A has an TextLayer so if you search words in an PDF-Reader it can find the words. What do you mean with annotated PDF ?

[ghost]

Like this output can we save it in PDF format?

[felixdittrich92]

Ah ok no not directly but you can do it on your own with our output:

from tempfile import TemporaryDirectory
import os

from PIL import Image
import cv2

from doctr.io import DocumentFile
from doctr.models import ocr_predictor

doc = DocumentFile.from_pdf("/home/felix/Desktop/Versicherungsbedingungen-08-2021.pdf")
predictor = ocr_predictor(reco_arch="crnn_mobilenet_v3_small", det_arch="db_mobilenet_v3_large", pretrained=True).cuda()
res = predictor(doc)
export = res.export()

with TemporaryDirectory() as tmpdir:

    for page in export["pages"]:
        page_idx = page["page_idx"]
        height, width = page["dimensions"]
        for block in page["blocks"]:
            for line in block["lines"]:
                for word in line["words"]:
                    xmin = int(word['geometry'][0][0] * width)
                    ymin = int(word['geometry'][0][1] * height)
                    xmax = int(word['geometry'][1][0] * width)
                    ymax = int(word['geometry'][1][1] * height)
                    cv2.rectangle(doc[page_idx], (xmin, ymin), (xmax, ymax), (255, 0, 0), 2)

        cv2.imwrite(os.path.join(tmpdir, f'{page_idx}.png'), cv2.cvtColor(doc[page_idx], cv2.COLOR_BGR2RGB))

    saved_files = [os.path.join(tmpdir, file_path) for file_path in sorted(os.listdir(tmpdir))]
    images = [Image.open(file_path) for file_path in saved_files]
    images[0].save('overlayed.pdf', save_all=True, append_images=images[1:])

[felixdittrich92]

If you need any explanation let me know 👍