I find that some of my archive photos are of large books which, when laid open, produce a curve in the lines of the document. Leo really seems to struggle with following a curved line, so it would be helpful if there was an option to manually show it where the lines are before transcription, rather than having to work out which lines it’s gotten mixed up by looking at the transcription.
I’m thinking of something similar to the line-tracing function when training a model in Transkribus, where you manually instruct the program to follow a line you’ve drawn - it would be great because sometimes correcting the transcription is almost impossible because it’s so confused.
Thank you for this Clare! We’re currently looking into ways that we can allow users to fine tune the model with their own training data. If you have any more notes on how you would use that functionality, please let us know. 
1 Like
I second this suggestion because I just noticed how I have accidentally taken some pictures that include two documents (i.e. one lying on top of another.) The transcription of the page in which I was interested was nearly perfect but then the other document lying beneath it rendered the full transcription somewhat difficult to decipher. I could crop my photos but sometimes that would cut out important sections of the pages I care most about. Overall, though, I was impressed that Leo worked so well with this difficult assignment.
1 Like