This repository provides the official PyTorch implementation of the following paper:

Contrastive Adversarial Learning for Person Independent Facial Emotion Recognition (AAAI 2021)

• Real-time demo with pre-trained weights

• Python (>=3.7)
• PyTorch (>=1.5.0)
• NumPy
• Wandb
• Fabulous (terminal color toolkit)
• Facenet-pytorch (face detection)

To install all dependencies, do this.

pip install -r requirements.txt

[21.06.04]: Add weights of encoder and FC layer.

[21.02.06]: UPLOAD full training and evaluation files.

[21.01.02] OPEN official pytorch version of CAL-FER.

1. Download three public benchmarks for training and evaluation (I cannot upload datasets due to the copyright issue).

• AffectNet

• Aff-Wild

• AFEW-VA

(For more details visit website)

2. Follow preprocessing rules for each dataset by referring pytorch official custom dataset tutorial.

1.Go to /src.

2.Train CAF-FER.

python main.py --gpus 0 --train 1 --freq 5 --csv_path <csv_path> --data_path <data_path> --save_path <save_path> --load_path <load_path>

3.Evaluate CAF-FER.

python main.py --gpus 0 --train 0 --csv_path <csv_path> --data_path <data_path> --load_path <load_path>

• Arguments
• gpus: GPU number (in case of single GPU, set to 0)
• train: 1 (training phase), 0 (evaluation phase)
• freq: Parameter saving frequency
• csv_path: Path to load name and label script.
• data_path: Path to load facial dataset.
• save_path: Path to save weights.
• load_path: Path to load pre-trained weights.

1. Go to /Real_demo.

2. Run main.py.

• Facial detection and AV domain FER functions are equipped.
• Before that, you have to train and save Encoder.t7 and FC_layer.t7.

Please cite our paper if you find our work useful for your research:

@inproceedings{kim2021contrastive,
title={Contrastive Adversarial Learning for Person Independent Facial Emotion Recognition},
author={Kim, Dae Ha and Song, Byung Cheol},
booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
volume={35},
number={7},
pages={5948--5956},
year={2021}
}