Many models on different languages of various epochs have been trained in Transkribus already. What we are still missing is a general model, which can recognise a wide range of modern handwritings. We have set ourselves the goal to provide such a model to our users in near future: a big and robust model, that can deal with most of modern handwritings.
Thinking about how to create the big amount of training data needed, crowdsourcing came to our mind and the idea for the readme2020 project was born. That’s where you and other Transkribus users come into play: please help us setting up the training data by providing a piece of handwritten text!
If you would like to contribute, the following four steps are everything, which is needed in order to do so:
- Take a blank sheet of paper and write something down by hand, this can be any kind of text and topic. Of course, you can also simply copy a text.
- Download the DocScan-app on your Android mobile phone (unfortunately not available for iPhones) and scan your writing, by following the instructions, shown on the screenshots on this page: https://readcoop.eu/readme2020/
- Upload the document to Transkribus
- You will get a confirmation-email with the link to our web-interface. Please add the transcription of your text there.
That’s already it.
As soon as there will be enough data, we from the READ-COOP-team will start the training of the modern handwriting model. We are super-curious about the outcome and are looking forward to creating something beneficial for many people together with you.
With that link you can read about all about the project-details: https://readcoop.eu/readme2020/
