This project employs an ensemble of two powerful Convolutional Neural Networks (CNNs), ResNext and Xception, to detect deepfake videos. Deepfake technology has been rapidly evolving, leading to an increase in the creation and distribution of misleading videos. This poses a severe threat to privacy, security, and the authenticity of information. To combat this, we have developed a Deepfake Detection system using Video Magnification techniques.
The approach taken in this project involves analyzing minute changes or irregularities in videos that are often imperceptible to the human eye but can be indicative of deepfake manipulation. Using the video magnification technique, we amplify these minute changes in color and motion to make them discernible. These amplified videos are then fed into an ensemble model of ResNext and Xception, powerful CNNs known for their exceptional performance in image classification tasks. By combining the strengths of these two models, we aim to increase the accuracy and robustness of our deepfake detection system.
The purpose of sharing this project is to provide an open-source tool for researchers, developers, and anyone interested in deepfake detection. We hope that this project serves as a step forward in the fight against misinformation and helps uphold the authenticity of digital content.