MHyEEG

Hypercomplex Multimodal Emotion Recognition from EEG and Peripheral Physiological Signals :performing_arts:

Official PyTorch repository for the papers:

1. Hypercomplex Multimodal Emotion Recognition from EEG and Peripheral Physiological Signals, ICASSPW 2023. [IEEEXplore][ArXiv Preprint]
2. Hierarchical Hypercomplex Network for Multimodal Emotion Recognition, MLSP 2024. [IEEEXplore][ArXiv Preprint]
3. PHemoNet: A Multimodal Network for Physiological Signals, RTSI 2024. [IEEEXplore][ArXiv Preprint]

Authors:

Eleonora Lopez, Eleonora Chiarantano, Eleonora Grassucci, Aurelio Uncini, and Danilo Comminiello from ISPAMM Lab 🏘️

📰 News

• [2025.05.15] Released pretrained weights 💣
• [2025.05.14] Updated code with H2 and PHemoNet models from MLSP and RTSI papers! 👩🏻‍💻
• [2024.07] Extension papers have been accepted at MLSP and RTSI 2024!
• [2023.11.11] Code is available for HyperFuseNet! 👩🏼‍💻
• [2023.04.14] The paper has been accepted for presentation at ICASSP workshop 2023 🎉!

Overview :blush:

📚 Papers & Models

| Model | Paper | Arousal F1 | Arousal Acc | Valence F1 | Valence Acc | Highlights | Weights | |----------------------|-------------------------------------------------------------------------------------------------------------------------------|------------|-------------|------------|-------------|------------|---------| | 🥇 H2 | MLSP 2024 [IEEEXplore][ArXiv] | 0.557 | 56.91 | 0.685 | 67.87 | Hierarchical model with PHC-based encoders in modality-specific domains, achieves best performance | Arousal - Valence | 🥈 PHemoNet | RTSI 2024 [IEEEXplore][ArXiv] | 0.401 | 42.54 | 0.505 | 50.77 | PHM-based encoders with modality-specifc domains and revised hypercomplex fusion module | Arousal - Valence | 🥉 HyperFuseNet | ICASSPW 2023 [IEEEXplore][ArXiv] | 0.397 | 41.56 | 0.436 | 44.30 | Introduces hypercomplex fusion module | Arousal - Valence

How to use :scream:

Install requirements

pip install -r requirements.txt

Data preprocessing

1. Download the data from the official website.

2. Preprocess the data: python data/preprocessing.py

   • This will create a folder for each subject with CSV files containing the preprocessed data and save everything inside args.save_path.

3. Create torch files with augmented and split data: python data/create_dataset.py

   • This performs data splitting and augmentation from the preprocessed data in step 2.
   • You can specify which label to consider by setting the parameter label_kind to either Arsl or Vlnc.
   • The data is saved as .pt files which are used for training.

Training

To reproduce the results, use the corresponding configuration file for each model and task:

• configs/h2.yml → H2 model
• configs/phemonet.yml → PHemoNet
• configs/hyperfusenet_arousal.yml → HyperFuseNet for valence
• configs/hyperfusenet_valence.yml → HyperFuseNet for arousal

Run training with:

python main.py --train_file_path /path/to/arsl_or_vlnc_train.pt --test_file_path /path/to/arsl_or_vlnc_test.pt --config configs/config.yml

To do a sweep (used in HyperFuseNet paper) run: python sweep.py

Experiments will be directly tracked on Weight&Biases.

Cite

Please cite our works if you found this repo useful 🫶

• H2 model:

@inproceedings{lopez2024hierarchical,
  title={Hierarchical hypercomplex network for multimodal emotion recognition},
  author={Lopez, Eleonora and Uncini, Aurelio and Comminiello, Danilo},
  booktitle={2024 IEEE 34th International Workshop on Machine Learning for Signal Processing (MLSP)},
  pages={1--6},
  year={2024},
  organization={IEEE}
}

• PHemoNet:

@inproceedings{lopez2024phemonet,
  title={PHemoNet: A Multimodal Network for Physiological Signals},
  author={Lopez, Eleonora and Uncini, Aurelio and Comminiello, Danilo},
  booktitle={2024 IEEE 8th Forum on Research and Technologies for Society and Industry Innovation (RTSI)},
  pages={260--264},
  year={2024},
  organization={IEEE}
}

• HyperFuseNet:

@inproceedings{lopez2023hypercomplex,
  title={Hypercomplex Multimodal Emotion Recognition from EEG and Peripheral Physiological Signals},
  author={Lopez, Eleonora and Chiarantano, Eleonora and Grassucci, Eleonora and Comminiello, Danilo},
  booktitle={2023 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW)},
  pages={1--5},
  year={2023},
  organization={IEEE}
}

Want more of hypercomplex models? :busts_in_silhouette:

Check out:

• Multi-view hypercomplex learning for breast cancer screening, under review at TMI, 2022 [Paper][GitHub]
• PHNNs: Lightweight neural networks via parameterized hypercomplex convolutions, IEEE Transactions on Neural Networks and Learning Systems, 2022 [Paper][GitHub].
• Hypercomplex Image-to-Image Translation, IJCNN, 2022 [Paper][GitHub]