Deepfake Detection with Transfer Learning

Overview

This project applies transfer learning to a deepfake image classification task by extracting pretrained ResNet18 embeddings and training a PyTorch multilayer perceptron on top of them.

Methods

• Feature extraction with pretrained ResNet18
• Multilayer perceptron (MLP) classifier
• Stratified train/validation split
• 5-fold cross-validation
• Hyperparameter tuning for learning rate and weight decay
• Test set prediction generation

Data

Image tensors for binary deepfake classification with separate training and test sets.

Tools

Python, PyTorch, torchvision, scikit-learn, NumPy, matplotlib