search

Grzego / handwriting-generation

Implementation of handwriting generation with use of recurrent neural networks in tensorflow. Based on Alex Graves paper (https://arxiv.org/abs/1308.0850).
MIT License
531 stars 107 forks source link

Are you able to generate handwriting for special symbols such as @,✓, +, $ and £? #23

Open AIMLAPP opened 4 years ago

AIMLAPP commented 4 years ago

Hello,

I've been exploring different handwriting generators and would like to ask a question about this one.

May I please ask if your model, given the data you used to train it, is able to generate special symbols such as @, +, $ and £, ✓, as well as uppercase Q and Z?

Many thanks! :)

Grzego commented 4 years ago

Hi @AIMLAPP, unfortunately all of those symbols where not present in training dataset (except + I think). This means that the generator will not be able to generate sequences with those special special characters (any of those characters would be replaced with <NULL> meaning it will be effectively ignored during generation).

I'm not sure about quality of uppercase letters (I think those are rather rare) but they are present in dataset so it should be possible to generate them. :)

AIMLAPP commented 4 years ago

Hi @Grzego,

I hope you're well.

Ah OK, I understand. Thank you very much for your reply, I really appreciate it. :)

I have an additional question I would like to ask please. In your "ReadMe" section, I see that you have used the On-Line dataset. The xml files contain the handwriting data. Do you know how to convert these into a .jpg/.png such that I can visualise the handwriting (i.e. convert the On-Line dataset into "Offline")? I'm thinking of creating my own dataset like the xml files, but would like to also visualise and verify them through this conversion.

Thank you and have a good weekend!

Grzego commented 4 years ago

@AIMLAPP so it seems there is a ready off-line dataset available at "data/lineImages-all.tar.gz". According to this page where you can download and example from it, it contains .tif images that could be converted into .jpg/.png if needed. Hope this helps. :)

EDIT: Sorry, I missed your part about visualizing own dataset. So in this case you can look at this short script I posted in other issue. The parts with plotting should be the most interesting in your case. :)