Official implementation of our paper:

Exploring the Boundaries of Semi-Supervised Facial Expression Recognition using In-Distribution, Out-of-Distribution, and Unconstrained Data
Shuvendu Roy, Ali Etemad
In IEEE Transactions on Affective Computing, 2024

Dataset

We used the following dataset

1. AffectNet
2. FER-13
3. RAF-DB
4. KDEF
5. DDCF
6. CelebA

Once the dataset is downloaded use the scripts in datasets/preprocessing to preprocess the dataset. The porcessed dataset structure should look like this:

dataset
├── train
│   ├── class_001
|   |      ├── 1.jpg
|   |      ├── 2.jpg
|   |      └── ...
│   ├── class_002
|   |      ├── 1.jpg
|   |      ├── 2.jpg
|   |      └── ...
│   └── ...
└── val
    ├── class_001
    |      ├── 1.jpg
    |      ├── 2.jpg
    |      └── ...
    ├── class_002
    |      ├── 1.jpg
    |      ├── 2.jpg
    |      └── ...
    └── ...

Run

Modify the config files in config/ directory if needed.

python [ALGO_NAME].py --c [CONFIG_FILE]

Acknowledgement

This codebase is build upon the following repositories:

• TorchSSL
• SemiCLS
• CoMatch

We thank the authors for releasing their code. If you use our model and code, please consider citing these works as well.