Awesome

Pytorch Implementation for "Minority-Oriented Vicinity Expansion with Attentive Aggregation for Video Long-Tailed Recognition" (AAAI 2023 Oral) and Imbalanced-MiniKinetics200 dataset.

Arxiv | Paper | Imbalanced-MiniKinetics200 | Project Page | Video

1. Requirements & Environments

To run the code, you need to install requirements below. We suggest to work with torch version (1.2 ~ 1.7.1). Other versions may work fine but we haven't tested them.

pip install -r requirements.txt

2. Training & Evaluation

Please refer to subdirectories for training each VideoLT and Imbalanced-MiniKinetics200 dataset.

Cite MOVE

If you find this repository useful, please use the following entry for citation.

@inproceedings{moon2023minority,
  title={Minority-Oriented Vicinity Expansion with Attentive Aggregation for Video Long-Tailed Recognition},
  author={Moon, WonJun and Seong, Hyun Seok and Heo, Jae-Pil},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={37},
  number={2},
  pages={1931--1939},
  year={2023}
}

Contributors and Contact

If there are any questions, feel free to contact with the authors: WonJun Moon and Hyun Seok Seong.

Acknowledgement

This repository is built based on VideoLT repository.