Train my own recognition model

[stealth0414]

i want to use english_g2.pth to train my own recognition model,is there any Tutorial? The deep-text-recognition-benchmark model looks like 200MB,It's a little big for me,thanks

[MdotO]

You may use this doc provided by author for your reference : https://github.com/JaidedAI/EasyOCR/blob/master/custom_model.md

[stealth0414]

Thank you, I have successfully done this part of the content. My datasets is mainly printing character. What's the size of datasets do you suggest to have better model?

[EivindKjosbakken]

The larger is typically better, as long as the dataset is diverse. When you have a lot of data samples, adding more will not be as impactful anymore, but this likely requires 10k plus images. If you want to quickly generate a large dataset, you can synthetically generate one, which I have written about in TowardsAI here https://pub.towardsai.net/how-to-make-a-synthesized-dataset-to-fine-tune-your-ocr-3573f1a7e08b if that is of interest. I hope this helps!