My UniMERNet

January 16, 2025 ยท View on GitHub

A cleaner and more organized version of UniMERNet.

Features

  • Maximizes code reuse through inheritance from Swin and MBart in transformers library, avoiding extensive code duplication
  • Use transformers-compatible interfaces (VisionEncoderDecoder, TrOCRProcessor, etc.)
  • Removes non-UniMERNET components

Usage

Convert a official model with convert.ipynb and then use it like transformers.VisionEncoderDecoderModel, just like TrOCR.

Credits

Besides direct Python imports, this project makes use of the following open-source projects:

  • tramsformers: State-of-the-art Machine Learning for Pytorch, TensorFlow, and JAX.
  • UniMERNet: A Universal Network for Real-World Mathematical Expression Recognition.

LICENSE

This repository is published under an MIT License. See LICENSE file.