OCR identify slash instead of number seven

[johnidm]

Issue description

In this image, the OCR identifies the slash as the number seven.

The result is: F 10724 LI J15 2331 H. The result should be F 10/24 L J15 2331 H

Is there any parameter or way to avoid this type of error?

This is a small bug of course, but I'm looking for a solution because this is common in my dataset

Code snippet to reproduce the bug

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

images_path = "image.png"
doc = DocumentFile.from_images(images_path)

model = ocr_predictor("db_resnet50", "crnn_vgg16_bn", pretrained=True)
result = model(doc)

page = result.pages[0]

words = []
for block in page.blocks:
    for line in block.lines:
        texts = [word.value for word in line.words]
        words.append(" ".join(texts))

print(" ".join(words))

Error traceback

N/A

Environment

Python 3.12.8
python-doctr 0.10.0
Ubuntu 22.04.5 LTS

Deep Learning backend

I'm using PyTorch, but same problem happens on TF as well.

[felixdittrich92]

Hi @johnidm 👋,

The easiest way to adjust the model to your needs see: https://mindee.github.io/doctr/using_doctr/custom_models_training.html
If the slash and seven mistake is the only issue it should be easy to fine tune the preferred model only on a few samples to your needs. :)

Best,
Felix

[johnidm]

Hi @felixdittrich92 thanks for your help.

Perfect, I'll study to train my own model to solve this problem.

Thanks so much.

[johnidm]

How to train text detection & recognition models for Vehicle Identification Number (VIN) extraction using Python & docTR