Tesseract retraining on tilted text

[Iknoor Singh]

Hi 

I want to train Tesseract on my custom data. The data contains a lot of text with tilted bounding boxes. Do we need to do some preprocessing to deskew the bounding boxes before training the Tesseract model? Or can we retrain the model directly with tilted text? 

For eg. if we have the coordinates of the bounding box as (left, bottom, right, top) and the box is tilted like in this case: Image Link  

Also, if we are only passing the left bottom and right top, how will the model know the orientation of the bounding box?

Ref https://stackoverflow.com/questions/63560975/tesseract-retraining-on-tilted-text

Thanks in advance.

Regards,

Iknoor