22/09/2025
Project Name
π Deepfake Detection using Transfer Learning (ResNet50)
Overview
In this project, I developed a deep learning model capable of detecting real vs. fake images. With the rise of AI-generated media, detecting manipulated or deepfake content is crucial for ensuring authenticity in digital platforms. This project leverages computer vision and deep learning techniques to classify images into real or fake with high accuracy.
Features
β
Preprocessing of image frames using OpenCV
β
Data handling and splitting with Pandas, NumPy, and Scikit-learn
β
Model training with PyTorch (ResNet50 backbone)
β
Training visualization using Matplotlib & Seaborn
β
Model interpretability with torchsummary & torchview
β
GPU support for faster training
Model Training
Base Model: ResNet50 (pretrained on ImageNet)
Approach: Transfer Learning
Custom Layers:
Fully Connected Layers (1024 β 512 β 1)
Batch Normalization & Dropout for regularization
LeakyReLU activations for better feature learning
Loss Function: Binary Cross-Entropy
Optimizer: Adam / SGD
Output: Sigmoid activation for binary classification
Results & Visualization
π Training and validation accuracy were tracked across epochs.
πΌοΈ Below are sample images from the dataset (real vs. fake) and model prediction visualizations.