Why text detector(DBNet) requires transcription information?

[moominsy]

Hello, I have a question regarding text detector(DBNet) model's training data.

Why does it need field "transcription" even though it is just text "detector"?
Since I only have a metadata of coordinates of text area, not transcription, I wonder if I can use dummy text such as "invalid" for transcription to train this model. I should not transcribe them as "###" since I don't want the model to ignore my data.

Thank you for your effort to support your amazing model.