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
520 stars 107 forks source link

How to changes the images data into xml files #10

Open hnn123 opened 5 years ago

hnn123 commented 5 years ago

Do you know how to changes the images data into xml files. I have the handwritting images data and labels, but I don't know how to change them into xml files.

Grzego commented 5 years ago

So you have images of handwriting and labels with text? I assume you want to generate those files to train a model that can then be used to generate handwriting. If that's the case, then converting images to handwriting data would already required a model able to generate handwriting (so this is almost impossible to do automatically, unless your images are similar enough to the those in IAM dataset).

Is your dataset of images and text labels available somewhere? I could look it up and say something more about this problem.

hnn123 commented 5 years ago

@Grzego I have images of handwriting and labels with text. Like the following picture image

image

These data are collected offline and almost not similar to the those in IAM dataset. After I look the detail of the data format of IAM dataset, I realized that it is almost impossible to convert my data into xml files. What I want to do is training a model that can be used to generate handwriting using my own data. Do you know other method which can do that .

Grzego commented 5 years ago

@hnn123 IIRC the IAM dataset has images, handwriting and text. So it could be possible to train model that uses image and text to predict handwriting. If this model trained well you could then transcribe your images. And if this transcribed data is good enough you could train final handwriting model only on those data.

Those are quite big ifs, so it probably will be hard to achieve.