Access Request for MER-PS Evaluation Data

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Access is limited to non-commercial scientific research. The files include the raw EEG/fNIRS signals and ground-truth valence-arousal labels used for the MER 2026 public leaderboard. Requesters must provide accurate Codabench team and account information, protect participant privacy, avoid re-identification and redistribution, and must not use these released labels to tune or submit models to an active MER 2026 leaderboard.

Log in or Sign Up to review the conditions and access this dataset content.

MER-PS Public Leaderboard Data

This release contains the four subject-disjoint test participants used for the public leaderboard of the MER 2026 MER-PS Physiological Signal-Based Emotion challenge. It is released for non-commercial research and reproducible evaluation.

Contents

MER_PS_public_leaderboard_hf/
├── data/
│   ├── predict1/
│   ├── predict2/
│   ├── predict3/
│   └── predict4/
├── annotations/
│   ├── predict1_label.mat
│   ├── predict2_label.mat
│   ├── predict3_label.mat
│   └── predict4_label.mat
├── targets.csv
├── sample_ids.csv
├── Targeted_emotions.txt
├── fNIRS_coordinates.csv
└── fNIRS_reservations.csv

Each data/<subject>/ directory contains synchronized EEG and fNIRS recordings:

data/<subject>/
├── EEG_baselines.mat
├── EEG_videos.mat
├── fNIRS_baselines.mat
└── fNIRS_videos.mat

Dynamic Labels

annotations/<subject>_label.mat follows the original REFED annotation layout. Each file contains video_1 through video_15. Each variable is an integer matrix of shape 2 x T sampled at 1 Hz:

  • row 1: valence
  • row 2: arousal

Labels use the original integer scale [1, 255]; (128, 128) is the center of the valence-arousal plane. targets.csv is the equivalent row-wise representation with columns sample_id, valence, and arousal.

The sample_id convention is predictN_VVV_TTT, where VVV is the one-based video index and TTT is the zero-based 1 Hz timestamp.

License

This dataset is available for non-commercial research under CC BY-NC-SA 4.0.

Downloads last month
39