Skip to content

[CVPR'22] DirecFormer: A Directed Attention in Transformer Approach to Robust Action Recognition

License

Notifications You must be signed in to change notification settings

uark-cviu/DirecFormer

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DirecFormer: A Directed Attention in Transformer Approach to Robust Action Recognition

This is the official implementation of the paper "DirecFormer: A Directed Attention in Transformer Approach to Robust Action Recognition".

DirecFormer

Installation

The installation of this repository is similar to the installation of SlowFast. The instruction can be found here

To prepare a dataset, you should follow the instructions here provided by SlowFast.

Testing

To test the model on the Jester dataset, you can perform the following commands:

python tools/run_net_tsm.py --cfg config/Jester/DirecFormer.yaml \
            TRAIN.ENABLE False \
            TEST.CHECKPOINT_FILE_PATH <PATH-TO-CHECKPOINT> \

Training and Optimization

Please contact the Project Investigator (Khoa Luu) for further information about training models, optimized models on-the-edge and low-cost devices.

Acknowledgements

This codebase is borrowed from SlowFast and TimeSFormer

Citation

If you find this code useful for your research, please consider citing:

@inproceedings{truong2021direcformer,
  title={DirecFormer: A Directed Attention in Transformer Approach to Robust Action Recognition},
  author={Truong, Thanh-Dat and Bui, Quoc-Huy and Duong, Chi Nhan and Seo, Han-Seok and Phung, Son Lam and Li, Xin and Luu, Khoa},
  booktitle={Computer Vision and Pattern Recognition},
  year={2022}
}

About

[CVPR'22] DirecFormer: A Directed Attention in Transformer Approach to Robust Action Recognition

Resources

License

Code of conduct

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages