Dulu

Dulu is a small python library for doing OCR on image. Under the hood it uses florence model from microsoft, which is a state of the art vision model for many downstream tasks including OCR.

Installation

pip install dulu

Example Code

Here is a snippet of the code used in the notebook:

from dulu import OCR
from PIL import Image

# Load the image
image = Image.open("./sample_image.png")
image

# Initialize the OCR object
ocr = OCR()

# Recognize text without bounding boxes
text = ocr.recognize_text("./sample_image.png", bbox=False)
print(text)

# Recognize text with bounding boxes
text_with_bbox = ocr.recognize_text("./sample_image.png")
print(text_with_bbox)

List of models

By default we use Florence-2-base-ft model, you can over ride this with any other above in the list when instantiating OCR(model_name="microsoft/{model_id})

Acknowledgments

• Florence2 model