Edouard Belval

Should be fixed by #174. Can you confirm?

If you find any open source Chinese fonts, I'd be willing to add them to the project.

Chinese script is not supported by the handwritten generative model. I'll add a more explicit explanation in the README.

It's surprisingly hard to imitate because the original dataset dates from 2012 and used a method that isn't really common nowadays. I would look into GAN-based character generation like [ScrabbleGAN](https://openaccess.thecvf.com/content_CVPR_2020/papers/Fogel_ScrabbleGAN_Semi-Supervised_Varying_Length_Handwritten_Text_Generation_CVPR_2020_paper.pdf)...

You can use `--name_format 2` but that will give you spaces instead of csv. You could just make a code change here: https://github.com/Belval/TextRecognitionDataGenerator/blob/master/trdg/run.py#L473

This project does not allow for affine/homography transformations. This is however a very good idea if you wish to make a PR. OpenCV has everything to do it, you could...

The current version does not return a mask for handwritten text. I am not currently working on it so feel free to try and implement it.

Your PyTorch dataset could hold the GeneratorFromStrings object and you can implement a lazy loading for the strings in your dataset. Since you dataset is only strings you could most...

I do agree that it ends up being a stupidly high value, the idea was just to keep the font size consistent so it wouldn't look terrible after up scaling....

This is caused but the `"` in the filename. You can fix it by using a different output format with `-na 2`. Here is the description of the format parameter:...