MicroVQA

This is the code for the MicroVQA benchmark (hosted on 🤗HuggingFace here) and the 🤖RefineBot method that removes language shortcuts from multiple-choice evaluations. They were published in the paper: MicroVQA: A Multimodal Reasoning Benchmark for Microscopy-Based Scientific Research.

The repo contains:

• eval evaluation code for the MicroVQA benchmark (hosted on 🤗HuggingFace here). See its README.
• refinebot is the 🤖RefineBot method for removing language shortcuts from MCQs. See its README.
• benchmark is code used in benchmark construction. See its README

If any of this is useful, please cite us!

@inproceedings{burgess2025microvqa,
  title={Microvqa: A multimodal reasoning benchmark for microscopy-based scientific research},
  author={Burgess, James and Nirschl, Jeffrey J and Bravo-S{\'a}nchez, Laura and Lozano, Alejandro and Gupte, Sanket Rajan and Galaz-Montoya, Jesus G and Zhang, Yuhui and Su, Yuchang and Bhowmik, Disha and Coman, Zachary and others},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={19552--19564},
  year={2025}
}