A comprehensive deep learning pipeline for detecting deepfake videos and images using state-of-the-art convolutional neural network architectures such as XceptionNet, MisoNet, ResNet, and VGG-19.
This project focuses on building and benchmarking deepfake detection models using large-scale datasets. It integrates both TensorFlow and TensorRT to accelerate inference and optimize model deployment for real-time performance.
- Trained on 80GB+ of real and deepfake media data.
- Achieved 71.5% average precision across multiple deep learning architectures.
- Implemented real-time detection using TensorRT.
- Evaluated robustness against adversarial examples.
- Published findings to advance ethical AI and forgery detection research. [Paper]
- Python, TensorFlow, Keras, TensorRT
- XceptionNet, MisoNet, ResNet, VGG-19
- NumPy, OpenCV, Matplotlib, Scikit-learn