Real vs Fake Image Classifier

This binary image classifier was developed for a private Kaggle competition to detect AI-generated (fake) faces. The model uses a convolutional neural network (CNN) implemented in PyTorch and was trained on 128×128 grayscale images. The dataset, composed of both real and fake faces, was split into training and validation sets, with preprocessing and augmentation applied to reduce overfitting.

The final model achieved 99% validation accuracy and outputs predictions to a CSV file for competition submission. Evaluation metrics such as precision, recall, and F1 score confirmed its strong performance across both classes.

Technologies & Concepts:
- PyTorch (CNN architecture, training loop)
- Jupyter Notebook for experiment tracking
- Data preprocessing, augmentation, and normalization
- Binary classification with dropout for regularization
- Evaluation with confusion matrix and F1 score

• See Project on Github