Skip to content

Pablo-Rimoldi/neural-classification-artificial-faces

Repository files navigation

Seeing the Uncanny: Neural Classification of Artificial Faces

Project Overview

This project investigates the neural representation of AI-generated hyper-realistic faces compared to real human faces. Despite behavioral difficulty in distinguishing them, our goal is to use Machine Learning and Deep Learning to classify brain responses (ERP) recorded via EEG.

Scientific Context

  • P.I.: Prof. Alice Mado Proverbio (Univ. Milano-Bicocca)
  • Supervision: Prof. Claudia Casellato (Univ. Pavia)
  • Team: Pablo Rimoldi, Tommaso Godino, Andrea De Paola, Giacomo Colombo

Technical Specifications

  • Data Source: 128-channel EEG (10/5% system), 512 Hz sampling rate.
  • Stimuli: 440 male/female faces (Real vs. GAN-generated).
  • Signal Processing:
    • Bandpass filter: 0.01-70 Hz.
    • Notch filter: 50 Hz.
    • Epochs: -100 ms to 800 ms.
    • Reference: Common Average Reference (CAR).

Classification Task

We aim to classify EEG trials into 4 categories:

  • Code 50: AI Male
  • Code 60: AI Female
  • Code 70: Real Male
  • Code 80: Real Female

Key Features

Focus on the 200-600 ms time window and specific electrodes: O1, O2, PO9, PO10, TP7, TP8, P3, P4, AF3, AF4, AFF1h, AFF2h, AFF3h, AFF4h.

Quick Start

  1. clone repository
   git clone https://github.com/Pablo-Rimoldi/neural-classification-artificial-faces
   cd neural-classification-artificial-faces
  1. Install dependencies:
    pip install -r requirements.txt
  1. Run cleaning:
    python src/data_cleaner.py

About

Neural classification of real vs. GAN-generated hyper-realistic faces using EEG ERP signals. Collaborative project for the Bachelor of AI (UNIPV/UNIMIB).

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors