EEG-based Motor Imagery empowered by Subject Identification in a Multi-task and Multi-scale framework

Simone Bove, Martino Giaquinto*, Member, IEEE, Gennaro Percannella, Member, IEEE, Alessia Saggese, Member, IEEE, Mario Vento, Member, IEEE

All the authors are with the Department of Information and Electrical Engineering and Applied Mathematics (DIEM), University of Salerno, Fisciano, 84084 Italy.

*Corresponding author: mgiaquinto@unisa.it

All the authors contributed equally to this work.

---

Official repository for the paper "EEG-based Motor Imagery empowered by Subject Identification in a Multi-task and Multi-scale framework", introducing MIRACLE, a multi-task, multi-scale deep learning architecture for EEG-based Motor Imagery (MI) classification jointly trained with a Subject Identification (SI) auxiliary task.

Note: This repository contains only the code required to test/evaluate the pretrained MIRACLE model. Pretrained model weights and the preprocessed test datasets are not included in this repository and must be downloaded separately (see Download below).

---

Table of Contents

• Overview
• MIRACLE Architecture
• Repository Structure
• Requirements
• Download
• Datasets
• Results
• Usage
• Citation
• Contact

---

Overview

MIRACLE is designed to improve Motor Imagery decoding from EEG signals by leveraging subject-specific information through a multi-task learning framework, combined with a multi-scale, frequency-selective feature extraction strategy.

This repository provides the test/inference pipeline used to reproduce the evaluation results reported in the paper on the:

• BCI Competition IV 2008 Dataset 2a
• BCI Competition IV 2008 Dataset 2b
• OpenBMI
• PhysioNetMI

MIRACLE Architecture

The figure below shows an overview of the proposed MIRACLE architecture, comprising:

• (a) Normalization block — based on Z-score standardization of the input EEG signals.
• (b) Frequency-Selective Multi-Scale (FSMS) block — four parallel convolutional branches paired with Squeeze-and-Excitation (SE) modules, designed to capture discriminative patterns across multiple frequency scales.
• (c) Feature-Selective Global Temporal (FSGT) block — integrates SE-based feature weighting with a Transformer encoder to model long-range temporal dependencies.
• (d) Multi-Task Classification (MTC) block — two classification heads, one for Motor Imagery (MI) classification and one for Subject Identification (SI), trained jointly to improve generalization and subject-invariant/subject-aware feature learning.

Figure: Overview of the proposed MIRACLE architecture — (a) normalization, (b) FSMS block, (c) FSGT block, (d) MTC block.

Repository Structure

.
├── README.md
├── requirements.txt
├── MIRACLE.py # MIRACLE model definition (FSMS, FSGT, MTC blocks)
├── test_motor_imagery.py # Test / inference entry point
├── preprocess_test_OpenBMI.py # Merges the split OpenBMI test set into a single .npz file
├── utils.py # Data loading, normalization, metrics, losses, network factory
├── Test_Sets/ # >>> downloaded test data goes here (flat .npz files, no subfolders) <<<
│ ├── test_2a.npz
│ ├── test_2b.npz
│ ├── test_OpenBMI.npz # created by preprocess_test_OpenBMI.py after merging the two downloaded parts
│ └── test_PhysioNetMI.npz
└── Weights_Models/ # >>> downloaded model weights archive extracts here, with these subfolders already included <<<
├── Results_2a/
│ ├── MIRACLE_mean_data.pt
│ ├── MIRACLE_std_data.pt
│ ├── MIRACLE_seed71_validation_log.txt
│ └── MIRACLE_seed71_best_model_fold{N}.pth
├── Results_2b/
│ ├── MIRACLE_mean_data.pt
│ ├── MIRACLE_std_data.pt
│ ├── MIRACLE_seed157_validation_log.txt
│ └── MIRACLE_seed157_best_model_fold{N}.pth
├── Results_OpenBMI/
│ ├── MIRACLE_mean_data.pt
│ ├── MIRACLE_std_data.pt
│ ├── MIRACLE_seed149_validation_log.txt
│ └── MIRACLE_seed149_best_model_fold{N}.pth
└── Results_PhysioNetMI/
├── MIRACLE_mean_data.pt
├── MIRACLE_std_data.pt
├── MIRACLE_seed131_validation_log.txt
└── MIRACLE_seed131_best_model_fold{N}.pth

Requirements

• Python 3.9
• PyTorch
• einops
• NumPy
• scikit-learn
• Matplotlib
• Seaborn

Install all dependencies with:

pip install -r requirements.txt

Download

Since the pretrained MIRACLE model weights and the preprocessed test datasets are not hosted in this repository, they must be downloaded from the corresponding release before running the test scripts:

• 🔗 Model weights release

• 🔗 Test sets release

• Model weights: distributed as a single archive that already contains the correct subfolder structure (Weights_Models/Results_2a/, Weights_Models/Results_2b/, Weights_Models/Results_OpenBMI/, Weights_Models/Results_PhysioNetMI/). Simply extract it in the repository root — no manual reorganization needed.

• Test datasets: distributed as individual .npz files. Create a Test_Sets/ folder in the repository root (if not already present) and place all downloaded test files directly inside it (flat, no subfolders):

Test_Sets/test_2a.npz
Test_Sets/test_2b.npz
Test_Sets/test_OpenBMI_first.npz
Test_Sets/test_OpenBMI_last.npz
Test_Sets/test_PhysioNetMI.npz

Note on the OpenBMI test set: since it exceeds 2GB, it is distributed split into two parts, test_OpenBMI_first.npz and test_OpenBMI_last.npz. After placing both inside Test_Sets/, merge them into a single test_OpenBMI.npz file by running:

python preprocess_test_OpenBMI.py

This concatenates the two parts into Test_Sets/test_OpenBMI.npz and automatically deletes the two original split files. All other test sets (2a, 2b, PhysioNetMI) are ready to use as downloaded, with no preprocessing needed.

Once everything is extracted and placed as described, the folder layout should match Repository Structure above.

Datasets

Evaluation is performed on the following public EEG Motor Imagery benchmarks:

• BCI Competition IV — Dataset 2a: 4-class Motor Imagery (left hand, right hand, feet, tongue), recorded from 9 subjects using 22 EEG channels.
• BCI Competition IV — Dataset 2b: 2-class Motor Imagery (left hand, right hand), recorded from 9 subjects using 3 EEG channels.
• OpenBMI: large-scale public EEG Motor Imagery dataset (Lee et al., 2019), 2-class Motor Imagery (left hand, right hand), recorded from 54 subjects using 62 EEG channels.
• PhysioNetMI: EEG Motor Movement/Imagery dataset (Schalk et al., 2004), 2-class Motor Imagery (left hand, right hand), evaluated here on 106 subjects using 64 EEG channels.

Only the test partitions required to reproduce the paper's results are provided via the download link above; the raw/original datasets are publicly available from the respective official sources for research purposes.

During training, the MI and SI task losses are combined with a weighting coefficient α, set to 0.001 for Dataset 2a, 0.01 for Dataset 2b, 0.001 for OpenBMI, and 0.1 for PhysioNetMI.

Results

The tables below report the per-subject classification accuracy (%) obtained on the test sets, across all the seeds evaluated for each dataset. For simplicity, only the model weights of the best-performing seed per dataset are made available for download (highlighted in bold below): seed 71 for Dataset 2a and seed 157 for Dataset 2b.

Dataset 2a (α = 0.001)

Seed	S1	S2	S3	S4	S5	S6	S7	S8	S9	Average
42	87.15	64.24	94.44	77.78	54.51	68.06	87.85	85.76	78.12	77.55
71 (best)	86.81	70.49	94.44	81.94	67.01	70.14	87.50	86.81	77.78	80.32
101	86.46	68.75	92.71	79.17	45.49	65.97	81.94	84.72	82.64	76.43
113	85.76	66.32	94.44	77.78	61.11	67.71	90.28	86.81	84.38	79.40
127	86.46	59.37	95.14	77.43	75.69	65.97	85.76	82.64	77.78	78.47
131	84.03	58.33	93.75	78.12	71.87	67.36	89.24	85.07	74.31	78.01
139	85.07	58.68	93.75	78.12	57.99	64.58	89.93	86.11	77.78	76.89
149	86.11	65.28	95.49	76.74	48.96	70.14	90.28	88.19	82.99	78.24
157	86.46	58.33	94.79	79.17	54.51	69.44	85.42	86.81	86.11	77.89
163	84.38	61.46	94.79	73.61	54.86	67.01	89.24	82.99	79.17	76.39
173	86.81	58.33	93.40	73.61	52.08	64.58	73.61	87.50	82.64	74.73
181	85.76	67.36	95.49	78.82	52.43	64.93	81.60	85.07	80.90	76.93
322	85.42	64.58	95.14	81.94	48.96	67.71	86.11	87.15	80.21	77.47
521	82.99	63.19	94.10	77.78	61.46	69.10	88.54	84.38	81.25	78.09
Average	85.69	63.19	94.42	78.00	57.64	67.34	86.24	85.72	80.43	77.63

Values are per-subject classification accuracy (%) on the MI task test set. The bolded row (seed 71) is the one whose model weights are provided via the Download link and used in the Usage example commands below.

Dataset 2b (α = 0.01)

Seed	S1	S2	S3	S4	S5	S6	S7	S8	S9	Average
42	78.75	73.21	82.19	93.44	97.50	85.00	91.87	96.56	87.19	87.30
71	78.12	72.14	82.50	94.69	99.06	82.19	91.87	96.25	89.06	87.32
101	79.37	72.14	84.38	95.00	98.44	88.75	91.87	96.88	87.19	88.22
113	78.12	72.50	83.44	96.25	98.44	88.12	92.81	95.63	86.56	87.99
127	79.37	71.07	82.81	94.38	97.50	85.31	92.50	95.94	88.44	87.48
131	77.81	70.36	82.19	95.63	96.88	86.87	91.25	95.94	86.87	87.09
139	75.00	71.43	83.75	89.69	97.81	87.50	93.75	96.25	88.12	87.03
149	78.44	69.64	84.69	95.94	98.12	83.44	91.56	95.31	87.81	87.22
157 (best)	79.37	74.29	86.25	93.44	97.50	89.06	92.19	97.19	89.69	88.78
163	78.44	72.86	85.31	94.69	98.12	85.62	91.25	96.25	87.50	87.78
173	79.06	71.07	82.19	95.63	97.81	84.06	92.50	95.94	88.13	87.38
181	77.50	71.43	84.38	93.44	98.12	88.13	92.81	95.94	87.81	87.73
322	76.56	72.14	85.31	95.00	98.12	87.50	92.19	96.56	89.06	88.05
521	81.25	73.57	85.00	95.63	97.81	86.25	92.81	95.31	87.81	88.38
Average	78.37	71.99	83.88	94.49	97.94	86.27	92.23	96.14	87.95	87.70

Values are per-subject classification accuracy (%) on the MI task test set. The bolded row (seed 157) is the one whose model weights are provided via the Download link and used in the Usage example commands below.

Dataset OpenBMI (α = 0.001)

Subject	42	71	101	113	127	131	139	149	157	163	173	181	322	521	Average
S1	87.00	84.50	85.00	87.00	83.00	85.50	86.00	83.50	85.50	84.50	81.00	85.00	86.00	82.00	84.68
S2	79.00	84.00	80.50	81.50	84.00	78.50	84.50	82.50	79.50	80.00	81.00	80.00	81.00	82.00	81.29
S3	98.50	98.00	97.00	98.50	97.00	96.00	98.50	99.00	99.50	97.50	97.00	98.00	98.00	97.50	97.86
S4	87.50	86.50	80.50	83.50	88.00	87.00	86.50	90.50	84.50	88.50	86.00	89.00	91.00	90.00	87.07
S5	86.00	89.50	89.50	88.00	92.00	87.50	92.00	90.00	87.50	85.00	87.50	86.50	88.50	89.00	88.46
S6	99.00	97.50	98.00	97.50	99.50	98.00	98.00	99.00	98.00	99.00	98.00	98.50	99.50	98.50	98.43
S7	83.50	85.00	80.00	85.00	86.50	83.50	86.00	85.00	84.50	86.50	86.00	88.00	84.00	87.50	85.07
S8	88.00	87.50	86.00	87.50	88.00	85.00	91.50	88.50	89.50	86.50	87.00	85.00	86.00	85.50	87.25
S9	83.50	82.50	85.50	82.50	82.50	82.50	85.00	82.00	85.50	82.50	82.50	82.00	84.00	84.50	83.36
S10	68.00	66.50	67.00	71.50	68.50	67.00	73.50	68.00	72.00	67.00	66.00	67.50	72.50	69.00	68.86
S11	73.00	73.00	73.00	72.50	73.00	73.00	72.00	75.50	72.50	74.00	76.00	73.00	73.00	73.50	73.36
S12	86.00	84.00	86.00	84.50	82.00	82.00	83.00	83.00	84.00	83.00	83.00	87.00	85.00	81.50	83.86
S13	77.00	78.50	71.50	70.50	73.50	78.50	70.50	79.50	77.00	72.00	73.00	74.50	71.00	74.50	74.39
S14	72.00	62.00	72.00	76.00	66.50	68.50	70.00	72.50	68.00	67.50	81.00	71.50	74.00	65.50	70.50
S15	91.00	85.00	89.00	87.50	86.00	90.00	83.50	89.50	87.00	90.00	84.50	88.00	89.00	88.50	87.75
S16	93.00	91.50	92.00	90.50	92.50	93.50	92.50	92.00	92.50	90.00	91.50	91.50	90.50	92.00	91.82
S17	75.00	70.00	75.50	71.50	74.00	71.00	76.00	72.50	71.00	71.50	69.00	67.00	69.50	68.50	71.57
S18	94.50	89.00	94.50	89.50	94.00	94.00	93.50	95.50	96.00	87.50	93.50	96.00	92.00	93.00	93.04
S19	84.50	82.50	88.50	82.00	88.00	84.50	86.00	86.00	85.00	85.00	86.00	88.50	84.50	83.50	85.32
S20	94.50	89.50	92.00	92.00	93.00	93.50	91.50	92.00	93.00	93.00	91.50	94.00	91.50	93.00	92.43
S21	99.00	99.50	100.00	99.00	100.00	100.00	100.00	100.00	99.50	99.50	98.50	98.50	100.00	100.00	99.54
S22	83.50	85.50	84.50	83.00	84.50	84.00	85.50	84.50	84.50	81.00	84.00	85.50	79.50	84.00	83.82
S23	75.50	74.00	76.00	76.00	76.00	69.50	71.50	67.50	74.00	79.00	74.50	74.00	73.00	70.00	73.61
S24	60.50	55.00	65.00	62.50	53.50	60.50	56.50	59.50	61.00	61.50	66.00	60.50	63.00	61.00	60.43
S25	99.00	99.00	99.50	98.50	98.00	99.50	97.50	98.00	99.00	99.00	99.50	99.00	99.00	99.00	98.82
S26	83.00	86.00	85.50	82.00	86.00	85.50	85.50	84.00	86.50	85.50	89.00	85.00	85.50	87.00	85.43
S27	94.50	93.50	93.50	93.00	86.50	94.00	91.00	94.50	92.50	94.00	95.00	92.50	93.50	94.00	93.00
S28	96.50	96.00	97.00	98.50	93.50	97.50	95.00	96.00	96.00	97.50	96.00	97.50	98.00	95.50	96.46
S29	86.50	90.50	85.00	88.00	92.00	86.00	87.50	90.00	87.00	87.50	87.50	89.50	86.50	88.50	88.00
S30	75.00	73.00	74.50	76.50	74.00	73.50	75.00	76.50	78.50	72.00	74.00	76.50	74.00	75.00	74.86
S31	87.00	86.00	87.00	85.00	85.50	86.50	87.00	86.00	83.00	84.50	86.50	85.50	85.00	88.00	85.89
S32	96.00	97.50	98.00	98.00	97.00	97.00	98.50	98.00	97.50	97.50	97.50	96.50	97.50	96.50	97.36
S33	98.50	98.50	98.00	97.50	98.50	98.00	98.00	99.00	98.00	98.00	98.00	98.50	99.00	98.50	98.29
S34	66.00	66.00	69.50	67.00	71.50	62.50	65.50	67.50	67.00	62.50	70.00	67.00	66.00	66.50	66.75
S35	91.50	95.00	93.00	93.50	92.00	92.50	94.50	95.50	95.00	93.50	93.50	93.50	93.00	93.50	93.54
S36	99.00	99.00	98.50	100.00	98.50	99.50	98.50	98.50	99.00	99.00	100.00	98.50	99.50	100.00	99.11
S37	97.00	97.00	95.50	96.50	97.00	93.00	98.00	98.00	96.50	96.00	97.50	96.00	96.50	98.50	96.64
S38	73.50	79.00	72.50	71.00	74.50	74.00	81.50	81.00	80.00	70.50	78.00	84.50	77.00	80.50	76.96
S39	89.00	90.00	87.50	88.50	86.50	90.00	90.00	90.00	88.50	87.00	88.00	89.00	90.00	89.50	88.82
S40	82.50	81.50	80.50	80.50	85.00	79.00	83.50	83.00	78.50	82.50	79.50	89.00	78.50	83.50	81.93
S41	75.00	77.50	77.50	79.00	79.00	76.00	82.50	80.50	79.00	78.00	82.50	80.50	80.50	82.50	79.29
S42	76.50	77.50	75.50	76.50	77.00	77.00	74.00	74.00	77.00	76.00	76.50	76.00	76.00	74.50	76.00
S43	89.50	87.50	86.50	87.00	89.50	90.00	89.00	93.50	87.50	88.00	84.00	90.00	88.50	90.50	88.64
S44	95.50	94.00	95.50	95.50	91.50	95.50	93.00	93.50	96.00	95.50	95.00	95.50	96.00	94.00	94.71
S45	90.00	93.00	91.00	92.50	90.50	93.50	91.50	89.50	90.00	92.00	88.50	92.00	92.50	91.50	91.29
S46	85.50	81.50	83.50	86.50	81.00	82.50	84.00	85.50	83.00	84.50	85.00	84.00	81.00	84.50	83.71
S47	88.00	91.00	92.50	91.00	91.50	93.00	93.00	91.50	92.00	89.50	94.00	96.50	87.00	93.50	91.71
S48	82.00	83.00	82.50	84.50	83.00	78.50	79.00	83.00	81.00	81.50	82.00	85.50	80.50	83.00	82.07
S49	89.50	92.50	86.50	85.50	88.50	87.50	89.50	93.50	88.50	90.00	83.50	89.50	90.00	89.00	88.82
S50	55.50	62.00	61.50	62.50	65.50	57.00	59.00	61.00	61.50	53.00	65.50	56.50	57.50	61.00	59.93
S51	83.50	83.50	84.00	83.00	84.00	81.00	84.50	82.50	86.00	84.50	83.50	82.00	85.00	82.50	83.54
S52	90.00	90.50	90.00	92.00	90.50	92.00	89.50	92.00	89.50	88.50	93.00	91.00	89.50	92.00	90.71
S53	79.50	80.50	79.00	82.00	80.00	79.50	79.50	76.00	79.00	78.50	80.50	78.00	82.00	84.00	79.86
S54	75.50	75.00	76.00	73.00	72.00	74.50	72.00	72.50	73.00	72.50	74.00	76.00	74.50	71.00	73.68
Average	84.98	84.76	84.91	84.89	84.91	84.42	85.19	85.58	85.12	84.26	85.20	85.56	84.91	85.21	84.99

Values are per-subject classification accuracy (%) on the MI task test set; subjects are listed as rows and seeds as columns (the opposite orientation of the BCI IV 2a/2b tables above, given the larger number of subjects). The bolded column (seed 149) is the one whose model weights are provided via the Download link and used in the Usage example commands below.

Dataset PhysioNetMI (α = 0.1)

Subject	42	71	101	113	127	131	139	149	157	163	173	181	322	521	Average
S1	92.86	86.61	86.61	86.61	86.61	86.61	86.61	86.61	86.61	86.61	92.86	92.86	86.61	86.61	87.95
S2	86.61	80.36	85.71	85.71	86.61	85.71	93.75	85.71	100.00	85.71	100.00	79.46	100.00	86.61	88.71
S3	80.36	74.11	74.11	67.86	74.11	80.36	74.11	86.61	67.86	74.11	80.36	80.36	74.11	80.36	76.34
S4	100.00	93.75	100.00	100.00	100.00	100.00	92.86	92.86	87.50	100.00	100.00	92.86	100.00	100.00	97.13
S5	54.46	66.96	54.46	54.46	60.71	67.86	66.96	48.21	66.96	54.46	54.46	48.21	55.36	54.46	57.71
S6	92.86	92.86	92.86	100.00	100.00	100.00	92.86	85.71	92.86	100.00	92.86	92.86	85.71	100.00	94.39
S7	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00
S8	87.50	87.50	93.75	87.50	93.75	93.75	93.75	93.75	93.75	87.50	87.50	93.75	86.61	87.50	90.56
S9	87.50	80.36	66.96	79.46	74.11	74.11	87.50	79.46	81.25	87.50	87.50	79.46	73.21	87.50	80.42
S10	73.21	66.07	86.61	66.07	73.21	73.21	79.46	66.07	80.36	73.21	80.36	72.32	80.36	86.61	75.51
S11	78.57	93.75	87.50	85.71	85.71	85.71	85.71	85.71	85.71	92.86	100.00	78.57	85.71	92.86	87.43
S12	80.36	86.61	92.86	86.61	86.61	92.86	92.86	92.86	92.86	92.86	86.61	86.61	80.36	86.61	88.40
S13	67.86	59.82	67.86	79.46	66.96	73.21	74.11	75.00	54.46	60.71	73.21	59.82	66.96	75.00	68.17
S14	80.36	74.11	86.61	73.21	67.86	86.61	86.61	86.61	87.50	93.75	74.11	93.75	85.71	93.75	83.61
S15	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00
S16	65.18	57.14	71.43	64.29	64.29	51.79	50.89	78.57	78.57	73.21	57.14	66.07	68.75	57.14	64.60
S17	92.86	92.86	100.00	85.71	85.71	92.86	100.00	100.00	92.86	92.86	85.71	86.61	92.86	85.71	91.90
S18	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00
S19	86.61	86.61	80.36	79.46	79.46	86.61	81.25	86.61	87.50	80.36	79.46	79.46	86.61	73.21	82.40
S20	92.86	92.86	92.86	100.00	92.86	92.86	100.00	92.86	92.86	100.00	92.86	85.71	92.86	92.86	93.88
S21	61.61	62.50	55.36	55.36	62.50	61.61	68.75	62.50	62.50	62.50	68.75	55.36	81.25	62.50	63.08
S22	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00
S23	85.71	85.71	92.86	85.71	86.61	78.57	78.57	80.36	80.36	86.61	85.71	85.71	85.71	86.61	84.63
S24	68.75	55.36	67.86	80.36	75.00	62.50	67.86	81.25	67.86	67.86	62.50	61.61	74.11	68.75	68.69
S25	87.50	55.36	50.00	75.00	68.75	68.75	75.00	62.50	62.50	62.50	56.25	80.36	81.25	68.75	68.18
S26	33.04	32.14	38.39	39.29	21.43	26.79	27.68	26.79	20.54	33.04	32.14	25.89	25.89	26.79	29.27
S27	80.36	74.11	73.21	66.96	74.11	66.96	73.21	74.11	74.11	66.96	74.11	73.21	79.46	79.46	73.60
S28	79.46	73.21	66.96	87.50	66.96	86.61	73.21	66.96	74.11	74.11	80.36	73.21	66.96	79.46	74.93
S29	85.71	85.71	92.86	85.71	85.71	85.71	85.71	85.71	85.71	86.61	85.71	85.71	85.71	92.86	86.80
S30	86.61	86.61	86.61	80.36	80.36	86.61	80.36	80.36	80.36	80.36	86.61	86.61	86.61	86.61	83.93
S31	93.75	93.75	86.61	87.50	86.61	93.75	87.50	93.75	93.75	93.75	93.75	93.75	80.36	86.61	90.37
S32	100.00	100.00	93.75	100.00	100.00	92.86	92.86	92.86	100.00	92.86	100.00	100.00	92.86	92.86	96.49
S33	93.75	75.00	93.75	93.75	93.75	100.00	87.50	100.00	100.00	100.00	100.00	100.00	93.75	93.75	94.64
S34	71.43	71.43	85.71	78.57	78.57	71.43	57.14	85.71	78.57	85.71	92.86	64.29	100.00	78.57	78.57
S35	78.57	78.57	78.57	78.57	78.57	78.57	78.57	71.43	78.57	85.71	85.71	78.57	78.57	78.57	79.08
S36	74.11	74.11	80.36	80.36	80.36	80.36	74.11	80.36	80.36	74.11	74.11	74.11	74.11	74.11	76.79
S37	92.86	92.86	92.86	71.43	92.86	85.71	92.86	85.71	85.71	57.14	85.71	92.86	71.43	85.71	84.69
S38	67.86	60.71	56.25	75.00	66.96	73.21	56.25	56.25	48.21	49.11	66.07	55.36	62.50	62.50	61.16
S39	59.82	81.25	81.25	66.96	80.36	81.25	81.25	81.25	80.36	81.25	66.96	81.25	66.96	81.25	76.53
S40	100.00	100.00	100.00	100.00	93.75	100.00	93.75	100.00	100.00	100.00	100.00	100.00	93.75	100.00	98.66
S41	100.00	100.00	100.00	93.75	93.75	93.75	93.75	93.75	93.75	100.00	93.75	100.00	100.00	100.00	96.88
S42	85.71	92.86	85.71	92.86	85.71	85.71	86.61	92.86	92.86	87.50	92.86	85.71	92.86	78.57	88.46
S43	86.61	79.46	78.57	86.61	85.71	86.61	79.46	93.75	72.32	72.32	93.75	79.46	93.75	79.46	83.42
S44	87.50	80.36	87.50	87.50	87.50	80.36	80.36	87.50	87.50	87.50	87.50	81.25	80.36	87.50	85.01
S45	54.46	60.71	60.71	61.61	60.71	60.71	60.71	67.86	67.86	67.86	60.71	61.61	60.71	60.71	61.92
S46	60.71	52.68	66.07	52.68	59.82	66.96	52.68	60.71	66.96	60.71	74.11	52.68	59.82	59.82	60.46
S47	85.71	92.86	85.71	86.61	86.61	81.25	81.25	75.00	75.00	87.50	79.46	86.61	79.46	93.75	84.06
S48	92.86	100.00	100.00	92.86	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	92.86	98.47
S49	80.36	80.36	86.61	100.00	87.50	80.36	86.61	100.00	92.86	86.61	93.75	87.50	80.36	100.00	88.78
S50	80.36	79.46	85.71	80.36	87.50	93.75	72.32	66.07	80.36	80.36	79.46	71.43	71.43	87.50	79.72
S51	64.29	71.43	78.57	78.57	86.61	92.86	79.46	100.00	85.71	92.86	78.57	92.86	66.96	92.86	82.97
S52	100.00	100.00	100.00	100.00	100.00	100.00	92.86	100.00	100.00	100.00	100.00	100.00	92.86	100.00	98.98
S53	93.75	100.00	100.00	100.00	93.75	100.00	100.00	100.00	93.75	93.75	100.00	100.00	100.00	93.75	97.77
S54	100.00	92.86	100.00	93.75	100.00	100.00	100.00	100.00	93.75	100.00	100.00	93.75	86.61	100.00	97.19
S55	86.61	86.61	86.61	92.86	86.61	92.86	86.61	92.86	92.86	92.86	86.61	92.86	86.61	92.86	89.73
S56	100.00	92.86	92.86	100.00	92.86	100.00	100.00	100.00	100.00	85.71	100.00	92.86	85.71	100.00	95.92
S57	93.75	93.75	93.75	93.75	93.75	100.00	93.75	93.75	100.00	93.75	93.75	93.75	93.75	93.75	94.64
S58	87.50	80.36	85.71	80.36	80.36	86.61	86.61	87.50	86.61	87.50	87.50	86.61	93.75	87.50	86.03
S59	59.82	51.79	58.93	73.21	52.68	52.68	66.07	52.68	58.93	66.07	58.93	66.07	65.18	66.07	60.65
S60	93.75	93.75	86.61	100.00	93.75	93.75	93.75	100.00	93.75	93.75	93.75	100.00	93.75	100.00	95.03
S61	100.00	100.00	100.00	92.86	100.00	92.86	92.86	92.86	92.86	100.00	92.86	100.00	92.86	100.00	96.43
S62	100.00	100.00	100.00	92.86	100.00	100.00	100.00	100.00	93.75	100.00	100.00	100.00	100.00	100.00	99.04
S63	58.93	66.96	65.18	73.21	73.21	73.21	66.07	65.18	58.93	73.21	74.11	73.21	65.18	65.18	67.98
S64	64.29	57.14	50.00	50.00	50.00	57.14	50.00	50.00	57.14	50.00	50.00	57.14	57.14	57.14	54.08
S65	81.25	81.25	87.50	87.50	87.50	87.50	87.50	87.50	87.50	80.36	87.50	87.50	86.61	87.50	86.03
S66	65.18	72.32	66.07	80.36	80.36	80.36	80.36	73.21	73.21	73.21	66.07	79.46	74.11	80.36	74.62
S67	52.68	51.79	64.29	64.29	57.14	64.29	64.29	64.29	64.29	57.14	64.29	39.29	50.89	64.29	58.80
S68	87.50	81.25	87.50	81.25	81.25	87.50	87.50	81.25	81.25	87.50	81.25	75.00	81.25	87.50	83.48
S69	79.46	93.75	100.00	93.75	92.86	93.75	86.61	93.75	86.61	92.86	79.46	85.71	78.57	79.46	88.33
S70	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00
S71	73.21	87.50	80.36	87.50	86.61	87.50	73.21	87.50	93.75	87.50	73.21	80.36	80.36	93.75	83.74
S72	50.89	58.04	65.18	79.46	79.46	72.32	72.32	66.07	72.32	73.21	79.46	79.46	72.32	72.32	70.92
S73	78.57	71.43	78.57	78.57	64.29	71.43	64.29	71.43	64.29	64.29	71.43	71.43	78.57	71.43	71.43
S74	71.43	85.71	78.57	71.43	85.71	78.57	71.43	85.71	92.86	71.43	71.43	64.29	78.57	78.57	77.55
S75	73.21	86.61	79.46	87.50	66.07	73.21	80.36	66.96	92.86	73.21	93.75	73.21	72.32	72.32	77.93
S76	71.43	78.57	71.43	78.57	85.71	78.57	71.43	85.71	85.71	71.43	78.57	78.57	85.71	78.57	78.57
S77	65.18	72.32	79.46	72.32	80.36	73.21	65.18	78.57	72.32	86.61	79.46	72.32	79.46	86.61	75.96
S78	73.21	67.86	66.96	66.96	60.71	74.11	66.96	67.86	66.96	72.32	66.96	79.46	80.36	80.36	70.79
S79	78.57	72.32	86.61	86.61	72.32	72.32	78.57	72.32	86.61	72.32	71.43	78.57	85.71	72.32	77.61
S80	93.75	92.86	87.50	100.00	100.00	100.00	93.75	87.50	87.50	86.61	93.75	86.61	100.00	93.75	93.11
S81	79.46	93.75	93.75	86.61	93.75	93.75	93.75	93.75	93.75	93.75	92.86	86.61	86.61	93.75	91.14
S82	86.61	86.61	86.61	86.61	80.36	92.86	86.61	92.86	86.61	85.71	86.61	86.61	86.61	79.46	86.48
S83	65.18	58.93	68.75	58.93	73.21	59.82	59.82	74.11	73.21	61.61	44.64	78.57	60.71	66.07	64.54
S84	87.50	68.75	75.00	81.25	81.25	87.50	75.00	75.00	75.00	75.00	87.50	87.50	67.86	93.75	79.85
S85	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	93.75	100.00	93.75	100.00	100.00	99.11
S86	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00
S87	52.68	46.43	41.07	53.57	52.68	53.57	47.32	33.93	54.46	49.11	46.43	47.32	41.07	40.18	47.13
S88	79.46	73.21	79.46	73.21	73.21	74.11	80.36	80.36	80.36	74.11	79.46	73.21	80.36	86.61	77.68
S89	93.75	100.00	93.75	93.75	100.00	100.00	93.75	100.00	92.86	100.00	100.00	100.00	92.86	93.75	96.75
S90	100.00	93.75	93.75	93.75	87.50	93.75	92.86	86.61	93.75	93.75	93.75	100.00	87.50	86.61	92.67
S91	72.32	78.57	78.57	72.32	72.32	72.32	72.32	78.57	78.57	78.57	78.57	72.32	72.32	78.57	75.44
S92	75.00	100.00	100.00	87.50	87.50	93.75	81.25	81.25	93.75	87.50	93.75	81.25	87.50	93.75	88.84
S93	33.04	50.89	44.64	51.79	46.43	39.29	44.64	39.29	51.79	45.54	53.57	47.32	59.82	38.39	46.17
S94	93.75	93.75	100.00	100.00	93.75	100.00	93.75	100.00	87.50	93.75	93.75	93.75	93.75	87.50	94.64
S95	93.75	87.50	93.75	87.50	79.46	80.36	79.46	93.75	86.61	87.50	87.50	93.75	93.75	86.61	87.95
S96	41.07	45.54	33.93	33.04	45.54	39.29	40.18	33.04	26.79	56.25	39.29	47.32	47.32	40.18	40.63
S97	72.32	87.50	73.21	87.50	87.50	73.21	66.07	81.25	73.21	52.68	79.46	80.36	81.25	59.82	75.38
S98	93.75	93.75	93.75	100.00	100.00	100.00	100.00	93.75	100.00	100.00	100.00	93.75	87.50	87.50	95.98
S99	71.43	64.29	71.43	64.29	71.43	71.43	64.29	71.43	64.29	71.43	71.43	64.29	64.29	71.43	68.37
S100	93.75	93.75	93.75	93.75	93.75	100.00	87.50	100.00	93.75	87.50	100.00	93.75	87.50	87.50	93.30
S101	80.36	74.11	74.11	73.21	67.86	74.11	74.11	80.36	80.36	79.46	80.36	80.36	74.11	74.11	76.21
S102	87.50	81.25	93.75	93.75	100.00	93.75	81.25	81.25	93.75	93.75	93.75	87.50	100.00	87.50	90.62
S103	73.21	66.96	72.32	72.32	66.96	73.21	73.21	73.21	66.96	66.07	66.96	59.82	66.07	59.82	68.36
S104	53.57	60.71	46.43	66.96	52.68	46.43	45.54	46.43	53.57	51.79	54.46	46.43	45.54	60.71	52.23
S105	80.36	86.61	80.36	80.36	80.36	86.61	86.61	80.36	74.11	86.61	86.61	74.11	92.86	74.11	82.15
S106	66.96	58.04	85.71	65.18	73.21	71.43	72.32	65.18	79.46	65.18	72.32	71.43	71.43	73.21	70.79
Average	80.57	80.33	81.85	82.08	81.43	82.47	80.27	81.78	81.92	81.28	82.40	80.80	81.09	82.09	81.45

Values are per-subject classification accuracy (%) on the MI task test set; subjects are listed as rows and seeds as columns (the opposite orientation of the BCI IV 2a/2b tables above, given the larger number of subjects). The bolded column (seed 131) is the one whose model weights are provided via the Download link and used in the Usage example commands below.

Usage

Once models and datasets have been downloaded and placed in the correct folders (remember to merge the OpenBMI test set first, see Download), run the test script specifying the test set, the saved-results folder, and the seed of the model weights to evaluate:

python test_motor_imagery.py \
--test_set Test_Sets/test_2a.npz \
--name_model MIRACLE \
--saved_path Weights_Models/Results_2a \
--seed 71 \
--paradigm Cross

python test_motor_imagery.py \
--test_set Test_Sets/test_2b.npz \
--name_model MIRACLE \
--saved_path Weights_Models/Results_2b \
--seed 157 \
--paradigm Cross

python test_motor_imagery.py \
--test_set Test_Sets/test_OpenBMI.npz \
--name_model MIRACLE \
--saved_path Weights_Models/Results_OpenBMI \
--seed 149 \
--paradigm Cross

python test_motor_imagery.py \
--test_set Test_Sets/test_PhysioNetMI.npz \
--name_model MIRACLE \
--saved_path Weights_Models/Results_PhysioNetMI \
--seed 131 \
--paradigm Cross

Results (F1, accuracy, balanced accuracy, and Cohen's kappa for both the MI and SI tasks) are saved to {saved_path}/Final_results_{name_model}_seed{seed}.json.

Note: This paper is currently under review. Citation details will be added once the work is accepted and published.

Citation

If you find this repository useful in your research, please consider citing our paper (citation will be added after publication).

---

Contact

For questions or further information, please contact the authors of the paper.