README.md

---
license: cc-by-nc-sa-4.0
language:
  - en
pretty_name: MER_PS

task_categories:
  - time-series-forecasting
  - tabular-regression
  - tabular-classification

tags:
  - eeg
  - fnirs
  - emotion-recognition
  - affective-computing
  - brain-computer-interface
  - multimodal
  - valence-arousal
  - regression
  - classification
  - competition

size_categories:
  - 10G<n<100G

extra_gated_prompt: "Access to MER_PS requires sharing your Hugging Face username and email address with the dataset authors. By requesting access, you agree not to use this dataset for experiments or applications that may cause harm to human subjects, and to use it only for non-commercial scientific research under the CC BY-NC-SA license."

extra_gated_fields:
  Full name : text
  Position : text
  Country : country
  Institution : text
  Department : text
  Supervisor name, if any : text
  Lab or supervisor homepage : text
  Research purpose : text
  I agree to use MER_PS only for non-commercial scientific research under the CC BY-NC-SA license :
    type: checkbox
    default: false
---
# MER 2026: MER-PS (Physiological Signal-Based Emotion) Codabench Public Training/Validation Data

This release contains the public MER_PS data for model development in the Codabench valence-arousal regression task. The subject identifiers are anonymized as `test_1` through `test_24`.

The task is to model continuous emotional state from synchronized EEG and fNIRS recordings. Dynamic labels are provided at 1 Hz for two affective dimensions:

- `valence`: pleasantness of the emotional state
- `arousal`: activation level of the emotional state

Valence and arousal use the original MER_PS raw label scale `[1, 255]`. The center of the valence-arousal plane is `(128, 128)`.

## Data Modalities

| Modality | Channels | Sampling Rate | Description |
|---|---:|---:|---|
| EEG | 64 | 200 Hz | EEG signals during baseline and video stimulation |
| fNIRS | 51 | 47.62 Hz | fNIRS signals during baseline and video stimulation |
| Annotation | 2 | 1 Hz | Continuous valence and arousal labels |

Each trial includes a 5-second baseline period before video onset. Each subject watched 15 emotion-inducing video clips.

## Directory Structure

```text
MER_PS_codabench_public_trainval/
├── README.md
├── Metadata.csv
├── SAM_score.csv
├── PANAS_score.csv
├── Targeted_emotions.txt
├── fNIRS_coordinates.csv
├── fNIRS_reservations.csv
├── data/
│   ├── test_1/
│   │   ├── EEG_baselines.mat
│   │   ├── EEG_videos.mat
│   │   ├── fNIRS_baselines.mat
│   │   └── fNIRS_videos.mat
│   ├── ...
│   └── test_24/
└── annotations/
    ├── test_1_label.mat
    ├── ...
    └── test_24_label.mat
```

## Files

`Metadata.csv` contains anonymized subject metadata.

`SAM_score.csv` contains post-trial subjective ratings based on the Self-Assessment Manikin scale.

`PANAS_score.csv` contains Positive and Negative Affect Schedule scores.

`Targeted_emotions.txt` lists the targeted emotion category for each video.

`fNIRS_coordinates.csv` contains the 3D coordinates of fNIRS channels.

`fNIRS_reservations.csv` contains fNIRS channel reservation records.

## Signal Files

Each subject folder under `data/` contains:

```text
EEG_baselines.mat
EEG_videos.mat
fNIRS_baselines.mat
fNIRS_videos.mat
```

The `.mat` files store trial-wise arrays such as `video_1`, `video_2`, ..., `video_15`.

EEG arrays are organized as:

```text
channel × time
```

fNIRS arrays are organized as:

```text
signal_type × channel × time
```

The fNIRS signal type dimension includes HbO, HbR, HbT, Abs 780 nm, Abs 805 nm, and Abs 830 nm.

## Annotation Files

Each file in `annotations/` contains dynamic valence-arousal labels for one anonymized subject:

```text
annotations/test_1_label.mat
```

Each annotation array has shape:

```text
2 × time
```

The first row is valence and the second row is arousal.