How to train custom model to recognize vertical as well as horizontal text?

[faisnain]

I have a document which is a combination of both horizontal as well as vertical text in it. In order to train a own model should the vertical text be given to the model in the same way as horizontal? eg: In labels.csv suppose we have the vertical word as "flamingo" in the image and its word in the labels.csv should be flamingo as well or in new line for each character? or the vertical image should be rotated and used for training to the model?