Crafting a Machine Learning Model for Discerning Synthetic and Authentic Images
Misha Shah, Gautam Jaiswal, Dolly Sharma · 2024
This paper shows the development of a Convolutional Neural Network (CNN) model with an attention mechanism and data augmentation for discerning synthetic and authentic images. Given the increase in AI-generated content there is a need for robust methods to identify an image's authenticity. The model employs ensemble learning by using ResNet50 and VGG16 architectures which are combined with a custom CNN. The project has detailed explanations of CNNs, data augmentation, TensorFlow, CUDA, ensemble models and the mathematical concepts underlying these techniques. With a dataset of $\mathbf{2 . 7}$ million images, the model aims for high accuracy in classification tasks.