"Universal" European text recognition model.

[Xargonus]

🚀 The feature

Train a model on a vocabulary that subsumes all of the similar European languages (German, French, Portuguese, Spanish, Czech, Polish, etc.) These languages only differ in a few letters.

Motivation, pitch

In all kinds of real-world texts, names from different languages appear, for example, a German text could contain French names.

Alternatives

No response

Additional context

No response

[felixdittrich92]

Hi @Xargonus ,

sounds really interesting 👍 The main question would be where to get data /lots of data (in best case real and balanced 😅 ). Feel free to share your ideas 😄 For your interest there is currently a wordgenerator implemented but it needs some improvements 👍 otherwise you can test trdg lib to generate some synthetic data

[felixdittrich92]

@Xargonus any update ? :)

[frgfm]

Hello @Xargonus :wave:

As suggested by Felix, the word generator of docTR can take in any character set. So if your interest is in visual recognition and not semantic understanding, I guess this is doable using these tools!

What do you think? If I misunderstood, feel free to elaborate your request :)

[felixdittrich92]

Hi @Xargonus 👋 any updates ?

[felixdittrich92]

Hi @Xargonus :wave:, would ask again if there are any updates from your side ? :)

[felixdittrich92]

Closing because no response