LaTeX OCR
LaTeX OCR
Install Performance
Overview
Convert an image containing a formula to corresponding LaTeX code.
There is a User Interface written with PyQt5
See here for performance information.
Install
As of right now you have to use a Python environment with PyTorch to use the model.
Easy installation:
pip install pix2tex[gui]
If you get an error, try to install PyTorch first.
For more information go to the GitHub repository.
A standalone version is planned. For Linux an AppImage is in development (PR).
Troubleshooting
If you run into problems, please see if you can find the solution in the issues-tab. If not open a new issue.
Contributing
I would love some help for further development of the project. If you want to contribute, there are many areas that need to be improved.
Handwritten formulae
Use CROHME, Im2Latex-Handwritten and maybe https://www.kaggle.com/aidapearson/ocr-data
Find better hyperparameters
I haven't tried out many sets of hyperparameters. Could potentially improve performance drastically.
Tweak model structure
Attention tricks, backbone model etc.
Distillation
Introduce more of the DeiT approach.
Data scraping
Fix equation extraction (regex) and collect more data.
Trace model
Using torchscript or ONNX. Useful for standalone application.
Desktop application
Never done anything with Qt. Help wanted!
If you want to help with any of the above or any other point, feel free to fork the repository and submit a pull request.