For the demonstration purpose, we're going to Import a fake video and check whether our application detects it. This is deepfake video of ‘The Shining,’ starring Jim Carrey [Link 🔗] (https://www.marketwatch.com/story/watch-this-deepfake-and-you-may-never-trust-a-video-recording-ever-again-2019-07-09)
We have obtained accelerated version of the model after using Intel® oneAPI Deep Neural Network Library
dem_video.mp4
This project aims to detect deepfake videos using advanced deep learning techniques, specifically leveraging the ResNeXt-50 model architecture. Deepfake videos, which are highly realistic but fabricated using AI algorithms, pose a significant threat to the authenticity of visual media content. By developing a robust framework for classifying videos as real or fake, we contribute to the ongoing efforts to combat the proliferation of misinformation and preserve the integrity of digital media platforms.
- Utilizes the Celeb-DF-v2 dataset, comprising a 6000+ diverse collection of real and synthesized videos featuring celebrity personas.
- Implements the ResNeXt-50 model architecture for video classification.
- Provides functionality for preprocessing videos, extracting frames, and applying data augmentation techniques.
- Offers inference capabilities for detecting deepfake videos in real time.
- Includes example scripts for training the model and evaluating its performance.
Leveraging the powerful resources provided by Intel Developer Cloud accelerated our development and deployment of the deepfake video detection model. We harnessed the computational capabilities of Intel's CPUs and XPUs to optimize critical components, such as video preprocessing, frame extraction, and model inference.
By taking advantage of Intel's high-performance computing infrastructure and optimized software libraries (e.g., oneDNN, Intel Distribution of OpenVINO), we significantly reduced the time required for data preprocessing, model training, and inference. This allowed for faster experimentation, iterative improvements, and ultimately, a more efficient deployment of our deepfake detection solution.
To run this project, you'll need the following libraries:
- Python 3.x
- oneDNN (Intel's Deep neural Network Library)
- PyTorch
- torchvision
- timm
- OpenCV
- PIL
- OneDNN
Contributions are welcome! Feel free to submit bug reports, feature requests, or pull requests to help improve this project.
Clone the project
git clone https://github.com/SaiNivedh26/Team-NooglersGo to the project directory
cd Team-NooglersInstall dependencies
pip install -r requirements.txtStart the server
npm run start- Developed a custom data pipeline for loading and preprocessing the Celeb-DF-v2 dataset. 📂
- Implemented the ResNeXt-50 model architecture using PyTorch for video classification. 🔥
- Utilized transfer learning techniques by fine-tuning the model on the Celeb-DF-v2 dataset. 🏋️♀️
- Leveraged Intel Developer Cloud's powerful computing resources and optimized libraries (e.g., oneDNN) to accelerate model training and inference. ⚡
This project is licensed under the MIT License.