search

sjvasquez / handwriting-synthesis

Handwriting Synthesis with RNNs ✏️
4.36k stars 592 forks source link

The current model does not support capital Q and Z #44

Open ohmycaptainnemo opened 4 years ago

ohmycaptainnemo commented 4 years ago

Hi,

When I give words with capital Z and Q in them, I get a complaint that the model does not have those letters as its trained letters. However, I noticed that your web demo has not issue with those letter. Could you please share your model that includes those characters?

JBisc commented 3 years ago

I checked the trainings data, and there is no capital Z and Q. Is there a Specific solution for that problem?

ohmycaptainnemo commented 3 years ago

Yeah I faced that issue too. The model used for the online demo seems to have Z and Q. I ended up using the online model. I believe the online demo stores the model on your pc (client side). So if you look at the code for the demo and find out what the model is called, you might be able to find the file on your pc. The second option, which is not elegant but works, is to use Selenium and python and save the generated results by the model. This works if you are only after data generation.

GirkovArpa commented 3 years ago

The online demo has been ported to rust, and a demo is available here.

bryandam commented 1 year ago

Sorry for necro-posting but I've been playing around with this and came up against the same issue. The currently online demo doesn't seem to have Q and Z and instead interprets Q as O and Z as either t or j depending on the style.

@ohmycaptainnemo: When you say training data, do you mean the data this project extracts from the IAM On-Line Handwriting Database or that the database itself lacks training for those characters?

@chavenor: I know you mentioned getting a Beam2 and trying to create your own dataset. Wondering how that went.

ohmycaptainnemo commented 1 year ago

@ohmycaptainnemo: When you say training data, do you mean the data this project extracts from the IAM On-Line Handwriting Database or that the database itself lacks training for those characters?

It has been a while and I do not exactly recall how I specifically came to the conclusion that the training data was lacking. Perhaps there was a particular error message or notification that let me to that conclusion. I am not sure (and prefer not to speculate) what the source of the issue is, but I think my observation was as per what I mentioned previously.

bryandam commented 1 year ago

Thanks @ohmycaptainnemo; upon re-read I see I missed two things: A) You were initially asking about and outright error. The project was throwing an error whereas the online demo 'worked'. Worked poorly mind you, but it works: image

B) That it was @JBisc who mentioned the training data. I downloaded the IAM database and yea, it seems to be missing these, characters. Which sorts of makes sense, they were writing real-world examples instead of pangrams and there's not a lot of sentences that are going to start with Q, X, and Z.

I did find the model binary file from the online demo (b.bin) but I'm not exactly sure how I would use that in this project. Not that it matters really, as even if I could it wouldn't lead to output that's legible. Someone would need to extend the IAM database to make these work.