Emotion Recognition on SEED

This is the second assignment of the course CS7327 at ShangHai Jiao Tong University, which focuses on transfer learning and domain adaptation approaches in emotion recognition tasks. We use all 15 subjects in SJTU Emotion EEG Dataset(SEED), for more information, please visit SEED Webpage.

In this assignment, our work are summarized as follows:

• We setup baseline with traditional machine learning approaches in tasks/baseline;
• We adopt domain adaptation neural networks with PyTorch implementation in tasks/DANN ;
• We apply vanilla transfer learning paradigm to this task with PyTorch implementation in tasks/vanilla_TL;

Setup environment

• First please set up environment in requirements.txt

Baseline

• To run a baseline model, run the script tasks/baseline/run.sh
• Hyper-parameters can be configured in tasks/baseline/conf/config.yaml

DANN

• To run a DANN model, run the script tasks/DANN/run.sh
• Hyper-parameters can be configured in tasks/DANN/conf/config.yaml

Vanilla TL

• To run pre-training, run the script tasks/vanilla_TL/run_backbone.sh
• To train a classifier, run the script tasks/vanilla_TL/run_classifier.sh
• Hyper-parameters can be configured in tasks/vanilla_TL/conf/config.yaml

Please note that the pre-trained checkpoints should by manually added to task/vanilla_TL/backbone_checkpoints, You should copy the checkpoints from hydra outputs.

DAN

• To run a DAN model, run the script tasks/DAN/run.sh
• Hyper-parameters can be configured in tasks/DAN/conf/config.yaml

MMD-AAE

• To run a MMD-AAE model, run the script tasks/MMD-AAE/run.sh
• Hyper-parameters can be configured in tasks/MMD-AAE/conf/config.yaml