This project is a real-time deepfake detection system implemented in PyTorch. Deepfakes are manipulated videos or images that use artificial intelligence to swap faces or modify visual content, often with malicious intent. The goal of this project is to develop a system capable of detecting deepfake videos in real-time.
Python 3.x PyTorch OpenCV (cv2) facenet-pytorch pytorch-grad-cam
Clone the repository to your local machine:
git clone https://github.com/Zhreyu/Realtime-Deepfake-Detection.gitChange directory to the cloned repository:
cd Realtime-Deepfake-DetectionInstall the required Python packages using pip:
pip install -r requirements.txtDownload the InceptionResnetV1 checkpoint file and place it in the root directory of the repository.
To run the real-time deepfake detection on the video, execute the following command:
python video_detection.pyThe program will process each frame of the video, detecting faces and applying the deepfake detection algorithm. The results will be displayed in a window showing the original video with bounding boxes around the detected faces and labels indicating whether they are real or deepfake. Press 'q' to exit the program.
The deepfake detection model uses the InceptionResnetV1 architecture pretrained on the VGGFace2 dataset. You can experiment with different models or custom architectures by modifying the code in deepfake_detection.py.
The face detection is performed using the Haar Cascade classifier and the MTCNN face detector. You can explore other face detection methods to see how they perform in this context.
The project is meant for educational and experimental purposes and should not be used for production-grade applications without further fine-tuning and validation.