Texotic

Texotic is a Python library to convert images of equations into LaTeX code based on the ONNXRuntime.

Cythonized fork of RapidLatexOCR. # TODO: finish this Modified fork of RapidLatexOCR.

• Works completely offline. Models are predownloaded
• Bug fixes:
  • Rewrote code using Numba:
    • TODO: document speed up
  • Removed usage of deprecated numpy methods
  • Stronger type checking via up-to-date type hint syntax, removed dependance on the inbuilt typing module
  • More comprehensive error handling and logging
  • More documentation and usage examples
  • Added more options for model inference
  • Refactored code

Rapid ⚡︎ Latex OCR

 

Introduction

• rapid_latex_ocr is a tool to convert formula images to latex format.
• The reasoning code in the repo is modified from LaTeX-OCR, the model has all been converted to ONNX format, and the reasoning code has been simplified, Inference is faster and easier to deploy.
• The repo only has codes based on ONNXRuntime or OpenVINO inference in onnx format, and does not contain training model codes. If you want to train your own model, please move to LaTeX-OCR.
• If it helps you, please give a little star ⭐ or sponsor a cup of coffee (click the link in Sponsor at the top of the page)
• Welcome all friends to actively contribute to make this tool better.
• ☆ Model Conversion Notes

Demo

TODO

• Rewrite LaTeX-OCR GUI version based on rapid_latex_ocr
• Add demo in the hugging face
• Integrate other better models
• Add support for OpenVINO

Installation

1. pip install rapid_latext_ocr library. Because packaging the model into the whl package exceeds the pypi limit (100M), the model needs to be downloaded separately.

   pip install rapid_latex_ocr

2. Download the model (Google Drive | Baidu NetDisk), when initializing, just specify the model path, see the next part for details.

   model name	size
   image_resizer.onnx	37.1M
   encoder.onnx	84.8M
   decoder.onnx	48.5M

Usage

• Used by python script:

  from rapid_latex_ocr import LatexOCR
  
  image_resizer_path = 'models/image_resizer.onnx'
  encoder_path = 'models/encoder.onnx'
  decoder_path = 'models/decoder.onnx'
  tokenizer_json = 'models/tokenizer.json'
  model = LatexOCR(image_resizer_path=image_resizer_path,
                  encoder_path=encoder_path,
                  decoder_path=decoder_path,
                  tokenizer_json=tokenizer_json)
  
  img_path = "tests/test_files/6.png"
  with open(img_path, "rb") as f:
      data = f. read()
  
  result, elapse = model(data)
  
  print(result)
  # {\frac{x^{2}}{a^{2}}}-{\frac{y^{2}}{b^{2}}}=1
  
  print(elapse)
  # 0.4131628000000003

• Used by command line.

  $ rapid_latex_ocr -h
  usage: rapid_latex_ocr [-h] [-img_resizer IMAGE_RESIZER_PATH]
                      [-encdoer ENCODER_PATH] [-decoder DECODER_PATH]
                      [-tokenizer TOKENIZER_JSON]
                      img_path
  
  positional arguments:
  img_path Only img path of the formula.
  
  optional arguments:
  -h, --help show this help message and exit
  -img_resizer IMAGE_RESIZER_PATH, --image_resizer_path IMAGE_RESIZER_PATH
  -encdoer ENCODER_PATH, --encoder_path ENCODER_PATH
  -decoder DECODER_PATH, --decoder_path DECODER_PATH
  -tokenizer TOKENIZER_JSON, --tokenizer_json TOKENIZER_JSON
  
  $ rapid_latex_ocr tests/test_files/6.png \
      -img_resizer models/image_resizer.onnx \
      -encoder models/encoder.onnx \
      -dedocer models/decoder.onnx \
      -tokenizer models/tokenizer.json
  # ('{\\frac{x^{2}}{a^{2}}}-{\\frac{y^{2}}{b^{2}}}=1', 0.47902780000000034)

Changlog

• 2023-09-13 v0.0.4 update:
  • Merge pr #5
  • Optim code
• 2023-07-15 v0.0.1 update:
  • First release

Code Contributors

Contributing

• Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.
• Please make sure to update tests as appropriate.

Sponsor

If you want to sponsor the project, you can directly click the Buy me a coffee image, please write a note (e.g. your github account name) to facilitate adding to the sponsorship list below.

License

This project is released under the MIT license.