Training a Sentence Text Recognition

[bely66]

Hi Everyone,
I wanted to discuss the issue of recognizing a varying number of words.
The current state of the models is that Detection cuts out every single word then the recognition model gives a prediction on the word level.

I wanted to explore the ability to Train a Text Recognizer that can recognize a sentence with a varying number of words, which would make the system more efficient in terms of speed given that we have the same accuracy as word level.

Does anyone have any idea on what recognition can work accurately with that?

Regards,