Attention Masking for Improved Near Out-of-Distribution Image Detection (BigComp2023, Oral)

"Attention Masking for Improved Near Out-of-Distribution Image Detection". BigComp2023 Oral.
Minho Sim, Jongwhoa Lee, and Ho-Jin Choi
Korea Advanced Institute of Science and Technology, Daejeon, South Korea

Overview

Dependencies

Install dependencies:

pip install -r requirements.txt

Setup project:

pip install .

Attention Masking

Visualization of the proposed method

Citation

If you find this repository useful, please consider citing:

@article{sim2023attention,
  title={Attention Masking for Improved Near Out-of-Distribution Image Detection},
  author={Sim, Minho and Lee, Jongwhoa and Choi, Ho-Jin},
  journal={BigComp},
  year={2023},
}

Acknowledgement

Our implementations of the vision transformer model and attention rollout algorithm are largely inspired by vit-pytorch and vit-explain.