RAPPG: Relation-Aware Progressive Pseudo-label Generation (implementation)

This repository contains the code used for the paper "RAPPG: Relation-Aware Progressive Pseudo-label Generation for Cross-subject MI-EEG Recognition."

The implementation is provided as reference for readers and researchers who want to understand or reproduce the method. It is an experimental, research-oriented codebase; a fully cleaned and reorganized release (a production-ready or well-packaged version) may be published in the future.

Files

• train.py — Main training / evaluation script for BCI datasets (e.g., BCI Competition IV 2a / 2b) using ATCNet + the RAPPG pseudo-labeling flow.
• train_hgd.py — Training / evaluation script for the HGD dataset.
• adaptation.py — Fine-tuning / adaptation script that applies progressive pseudo-labels to the target domain.
• atcnet_model.py — ATCNet model implementation (feature extractor + classifier head).
• ot_pseudolabel.py — Implementation of the relation-aware progressive pseudo-label generation (OT-based components and helpers).
• eegnet_model.py — Referenced by several scripts but not included in this repository; provide your own EEGNet implementation or remove these imports if not needed.
• datasets/ — Expected location for dataset .mat files used by the scripts.
• ckpts/ — Directory used to save trained model checkpoints.

Paper

RAPPG: Relation-Aware Progressive Pseudo-label Generation for Cross-subject MI-EEG Recognition

Dependencies

The code imports the following third-party packages (please install appropriate versions for your environment):

• Python 3.7+
• numpy
• scipy
• torch (PyTorch)
• einops
• pot (Python Optimal Transport; imported as ot in the code)

Standard library modules used include time and warnings.

Known issues / TODO

• The codebase is organized as experimental research code. Paths and interfaces are currently hard-coded in places and may be refactored in the future.