dntai / KERC21Baseline

3rd Korean Emotion Recognition Challenge, 2021

Description

Develop the emotion recognition system through multimodal analysis of various physiological signals

Url: https://www.kaggle.com/c/kerc2021/

Team information:

• Team Name: ADLER
• Team Members:
  • Nhu-Tai Do, [e-mail]
  • Tram-Tran Nguyen Quynh, [e-mail]
• Affiliation: Chonnam National University, South Korea

Baseline

Setup Dataset

• mnt/KERC21Dataset : link to KERC21 Dataset Directory. Example: ../../ssd_data/KERC21Dataset/
• data/KERC21Dataset: link to ../mnt/KERC21Dataset
• Structure of KERC21 Dataset

data/KERC21Dataset
├── train_labels.csv
├── train_personality.csv
├── val_personality.csv
├── test_personality.csv
├── train
│   ├── train_person_1
│   ├── train_person_2
│   └── train_person_n
│       ├── BVP_64.csv
│       ├── EDA_4.csv
│       ├── EEG_256.csv
│       └── TEMP_4.csv
├── valid
│   ├── valid_person_1
│   └── ...
└── test
    ├── test_person_1
    └── ...

How to Run BaseLine

• Run run_preprocessing_data.ipynb to extract train, valid, test features
• Configure configs/config.ini for training/testing configuration
• Run run_train_baseline.ipynb to train on the training data
• Run run_test_baseline.ipynb to test on the validation and testing data

Baseline Model for KERC2021 (From Organizer)

This code is distributed as a reference baseline for KERC2021 Emotion Recognition Challenge. This baseline is provided as an example for handling the dataset used in the competition.

Baseline Model performs the task of classification of Emotion into 4 quadrants of Arousal-Valence space, namely,

• HAHV: High Arousal, High Valence
• HALV: High Arousal, Low Valence
• LALV: Low Arousal, Low Valence
• LAHV: Low Arousal, High Valence

Usage

preprocessing.py extracts the following features from EEG and EDA and BVP signals for train, val, or test.

python preprocessing.py -r <PATH TO KERC21Dataset folder> -t train

• EEG (294 features)
  • mean, std, power frequency, wavelet coef. energy, spectral entropy, kurtosis, hjroth_complexity, p2p, fd, psd
• EDA (5 features)
  • 5 statistical features using PyTEAP package
• BVP
  • 17 statistical features using PyTEAP package

main.py run baseline model. - Baseline was trained on a 'training' set with 690 samples and hyperparameters were selected based on validation data. However, validation data is not available to the participants. Therefore, the baseline code only includes the training part.

generate_submission_csv.py to generate output on validation or test set - generate submission CSV files.

REQUIREMENTS

The baseline is implemented in PyTorch and depends on several external packages mentioned in requirements.txt

• mne==0.23.0
• mne-features==0.1
• numpy==1.19.5
• PyTEAP==0.1.2
• pandas==1.2.3
• scikit-learn==0.24.2
• scipy==1.6.3
• torch==1.8.1+cu111
• tqdm==4.60.0

Dataset

Dataset is divided into Training, Validation, and Test sets. However, only training data with labels and validation data is

Dataset Description

Dataset consists of following Modalities

• EEG recordings
  • 14 channels
  • 256 Hz
  • 60s duration
• EDA
  • 4Hz
  • 60s duration
• BVP
  • 64Hz
  • 60s duration
• Temperature
  • 4Hz
  • 60s duration
• Personality Traits
  • 5 personality Traits scored Low, Low-Mid, Mid-High and High