The codes are based on VideoMAE V2. We express our respect for their outstanding work. To prepare the environment, please follow the following instructions.
## install pytorch
conda install pytorch==1.13.0 torchvision==0.14.0 torchaudio==0.13.0 pytorch-cuda=11.6 -c pytorch -c nvidia
## install other requirements
pip install -r requirements.txt
The used datasets are provided in VideoEmotion-8 and Ekman-6. The train/test splits in both two datasets follow the official procedure. To prepare the data, you can refer to VideoMAE V2 for a general guideline.
We now provide the model weights in the following link.
You can easily evaluate the model by running the script below.
bash eval.sh
To get the fusion result, please run the test.py.
Please email to jinchow@gnnu.edu.cn if you are interested in our work. We would be fairly pleased if you find the code useful and cite the following paper.
@article{yi2024emotion,
title={Emotion recognition in user-generated videos with long-range correlation-aware network},
author={Yi, Yun and Zhou, Jin and Wang, Hanli and Tang, Pengjie and Wang, Min},
journal={IET Image Processing},
year={2024},
publisher={Wiley Online Library}
}