HiBug

This repo is the implementation of HiBug: On Human-Interpretable Model Debug. HiBug is a data centric tool for investigating bugs in visual deep learning, such as rare cases in dataset, model's correlation on visual feature.

Features

• Utilizes large language models and vision-language models for bug discovery and repair.
• Provides a human-understandable debugging process.
• Equipped with a user interface.

Installation

1. Clone the repository.
2. Set up the environment.

conda create -n HiBug python=3.7
conda activate HiBug
pip install -r requirements.txt

Getting Started

1. Get a ChatGPT API key. (Optional)
2. Follow the steps in run.ipynb.

Example

We provide a use case on celebA dataset in Code

Citing

If you find this repository useful for your work, please consider citing it as follows:

@inproceedings{chen2023hibug,
  title={HiBug: On Human-Interpretable Model Debug},
  author={Chen, Muxi and Li, Yu and Xu, Qiang},
  booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
  year={2023}
}