Authors: Milán András Fodor, Ivan Volosyak
Paper: “Towards Visual‑Fatigue‑Free BCI with Imperceptible Visual Evoked Potentials (I‑VEP)” (SMC 2025)

This repository contains all code, data‑processing scripts, and instructions to reproduce the two main pipelines in our I‑VEP study:

1. Nested CV + Optuna + CNN: I-cVEP_ResNet-CNN_NCV.py
   Runs a 4‑fold outer nested cross‑validation over all CSVs, using Optuna to tune:

• filter band centers & bandwidths
• window length & label shift
• CNN architecture (kernel size, filters, dropout, etc.)
• learning rate, weight decay, batch size, early stopping
  Outputs per‑fold .h5 models, .json hyperparameters, and a summary results.csv.

2. Leave‑One‑Trial‑Out SVC baseline: I-cVEP_SVC_decodability_check.py Implements a subject‑agnostic SVM (offline) baseline with causal Hilbert‑based features:

• Trial segmentation based on m-sequnce labels
• Band‑power (env + harmonics), low‑freq, and flip‑potential features
• Leave‑One‑Trial‑Out cross‑validation
  Prints per‑file LOTO and overall bit‑accuracy, with an optional verbose mode.

If you use this code, please cite: Fodor, M. A., & Volosyak, I. (2025). Towards Visual‑Fatigue‑Free BCI with Imperceptible Visual Evoked Potentials (I‑VEP). In Proceedings of IEEE SMC 2025. DOI: tba