Facial Prior Guided Micro-Expression Generation

Implementation of FOMM with AWP and MRAA with AWP in Facial Prior Guided Micro-Expression Generation (IEEE TIP 2024).

0. Table of Contents

• 0. Table of Contents

• 1. Visualizations

• 2. Run the Code

• 3. Citation

1. Visualizations

CASE	Driving	FOMM	FOMM w/ EWP	FOMM w/ AWP	MRAA	MRAA w/ AWP
1
2
3
4
5

• To facilitate micro-expression observations and reduce the storage space of this repository, all Gifs have been slowed down and compressed.
• Case 1-3: Positive. Case 4: Negative. Case 5: Surprise.

2. Run the Code

1. Procedures of running FOMM_with_AWP and MRAA_with_AWP are the same.

2. Prepare your dataset. CASME2, SAMM, and SMIC-HS are recommended. See related preparation instructions in this repository.

   • Divide into your_dataset/train and your_dataset/test

   • Create or modify yaml format file your_dataset_train.yaml in ./config

3. Train

   python run.py --config config/your_dataset_train.yaml

   Log, parameters and checkpoints would be saved in ./log

4. Test

   Create or modify csv format file your_dataset_test.csv in ./data

   python run.py --config config/my_dataset_test.yaml --mode animate --checkpoint path/to/checkpoint

   Generated videos would be saved in path/to/checkpoint/animation

3. Citation

If you find this work helpful in your research, please consider citing:

@ARTICLE{zhang2024tipfacial,
   author={Zhang, Yi and Xu, Xinhua and Zhao, Youjun and Wen, Yuhang and Tang, Zixuan and Liu, Mengyuan},
   journal={IEEE Transactions on Image Processing}, 
   title={Facial Prior Guided Micro-Expression Generation}, 
   year={2024},
   volume={33},
   number={},
   pages={525-540},
   doi={10.1109/TIP.2023.3345177}
}

@inproceedings{zhang2021acmmmfacial,
   author = {Zhang, Yi and Zhao, Youjun and Wen, Yuhang and Tang, Zixuan and Xu, Xinhua and Liu, Mengyuan},
   title = {Facial Prior Based First Order Motion Model for Micro-expression Generation},
   year = {2021},
   booktitle = {Proceedings of the 29th ACM International Conference on Multimedia},
   pages = {4755–4759},
   numpages = {5},
   series = {MM '21},
   doi = {10.1145/3474085.3479211}
}