JaidedAI/EasyOCR

Fine-Tuning EasyOCR on Urdu Textbooks

mujeeb-merwat opened this issue · 0 comments

I'm fine-tuning the EasyOCR model (arabic.pth) for Urdu TextBooks text recognition. Currently, I'm passing the image of an entire page as input and providing the corresponding text for the entire page. However, the performance is almost zero.

My question is: Is this approach correct, or should I split the page (image) into smaller regions for better performance?