A Simple Debiasing Framework for Out-of-Distribution Detection in Human Action Recognition (ECAI2023)

"A Simple Debiasing Framework for Out-of-Distribution Detection in Human Action Recognition". ECAI2023.
Minho Sim, Young-Jun Lee, Donkun Lee, 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 .

Train & Test

Train

Please refer to here for preparing ViViT model.

python -m finetune_vivit <<training parameters>>

Results

Case study

Citation

If you find this repository useful, please consider citing:

@article{sim2023simple,
  title={A Simple Debiasing Framework for Out-of-Distribution Detection in Human Action Recognition},
  author={Sim, Minho and Lee, Young-Jun and Lee, Donkun and Lee, Jongwhoa and Choi, Ho-Jin},
  journal={ECAI},
  year={2023},
}

Acknowledgement

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