diversen/doctr-torch-crnn_vgg16_bn-danish-v1

<p align="center"> <img src="https://doctr-static.mindee.com/models?id=v0.3.1/Logo_doctr.gif&src=0" width="60%"> </p>

Optical Character Recognition made seamless & accessible to anyone, powered by TensorFlow 2 & PyTorch

Task: recognition

https://github.com/mindee/doctr

This model does a good job if you need to do OCR on Danish documents.

Example usage:

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

reco_arch = from_hub('diversen/doctr-torch-crnn_vgg16_bn-danish-v1')
det_arch = "db_resnet50"

model = ocr_predictor(det_arch=det_arch, reco_arch=reco_arch, pretrained=True)
image = DocumentFile.from_images(['test.jpg'])

result = model(image)
result.show()

output = result.export()
text_str = ""

for block in output["pages"][0]["blocks"]:
    block_txt = ""
    for line in block["lines"]:
        line_txt = ""
        for word in line["words"]:
            line_txt += word["value"] + " "
        block_txt += line_txt + "\n"
    text_str += block_txt + "\n"

print(text_str)

Run Configuration

{ "arch": "crnnvgg16bn", "trainpath": "train-data", "valpath": "validation-data", "trainsamples": 1000, "valsamples": 20, "font": "FreeMono.ttf,FreeSans.ttf,FreeSerif.ttf", "minchars": 1, "maxchars": 32, "name": "doctr-torch-crnnvgg16bn-danish-v1", "epochs": 1, "batchsize": 64, "device": 0, "inputsize": 32, "lr": 0.001, "weightdecay": 0, "workers": 16, "resume": "crnnvgg16bn20240317-095746.pt", "vocab": "danish", "testonly": false, "freezebackbone": false, "showsamples": false, "wb": false, "pushtohub": true, "pretrained": true, "sched": "cosine", "amp": false, "findlr": false, "earlystop": false, "earlystopepochs": 5, "earlystop_delta": 0.01 }