Deep Fake Image Classification Engine Using Inception-ResNet-V1 Network

The goal of this project is to develop a real or fake facial image classification system using deep learning techniques. The main focus is on the InceptionResnetVl model pretrained on the VGGFace2 dataset and Kaggle DeepFake Classification Dataset for face classification. The application uses the facenet_pytorch library to perform face detection and preprocessing. Main features of the project include predicting the authenticity of a given facial image as either “real” or “fake”. This is achieved by implementing the InceptionResnetVl model, which is fine-tuned for binary classification with a single output representing the probability of authenticity. The system is designed to run on the GPU when available, enabling faster computing. Grad-CAM (gradient-weighted class activation mapping) technique is used to ensure the explainability of classification decisions. This method creates class activation maps that visually highlight the regions of the input image that influenced the classification decision. The Grad-CAM printout is then superimposed on the original facial image, creating an interpretable visualization of the model's decision-making process. The Gradio user interface is used for the interactive presentation of the project. Users can upload their face and get real vs. false predictions and visual explanations generated by the Grad-CAM algorithm. The user interface also displays a confidence score for each forecast, giving users an idea of how reliable the model is. In general, this project provides a comprehensive system to classify real and fake faces, and the proposed method reaches the accuracy of 97% both in training and testing.