Evgenii Dolotov

I see what you mean. One approach is to find a principal direction of a slant and measure its slope. Then, one could construct a simple rotation matrix and use...

This is a known problem. Sometimes, the model goes crazy after sampling certain number of points. Also, the model seems to struggle with producing rare letters and words. It is...

I have relaxed pinned versions for dependencies in the last commit: https://github.com/X-rayLaser/pytorch-handwriting-synthesis-toolkit/commit/b36353d84851e9776b9ccbf9d4f74ca0cfcbf83c Try to pull the changes from the main branch (or just copy paste the requirements file the way...

Yes, it is quite possibly the reason. I vaguely recall that in the code the attention window slides from left to right. If that's the case then, neural net might...

By the way, could you show a couple of typical samples generated by the neural net? It's just that there may be other issues with code besides the neural side...

Yep, try that. It might be the quickest fix.

Yes, I think it should be possible. Securing a large dataset continues to be the primary challenge. Your dataset must contain diverse and well-structured handwriting samples, presented as stroke sequences,...

If you already possess data with the necessary representation, implementing a custom data provider class should suffice. Essentially, this class must have two methods - get_training_data and get_validation_data. These methods...

Yes. I think it is possible, but I need to look into code and the paper to be certain. I vaguely recall that there is a technique that allows to...

From the usage point of view, one way to support this feature is the following: - add a boolean flag called "--export_style" to the synthesize.py command; when the flag is...