Will post the streamlit link as soon as possible
Inspiration
The inspiration behind Cipher-Vision stems from the growing concern surrounding deepfake technology and its potential misuse. With the rapid advancements in artificial intelligence, there is a need for robust tools to detect and combat the spread of manipulated media content. Cipher-Vision aims to contribute to this effort by providing an advanced deepfake detection solution focused on both images and, soon, videos.
What it does
Cipher-Vision is a deepfake detection application designed to analyze and identify manipulated content in images. Leveraging facial recognition technology and Grad-CAM visualization, the model can provide insights into its decision-making process. Users can upload images, and the application displays the model's prediction along with an interactive Grad-CAM visualization, highlighting regions influencing the prediction.
How we built it
The application is built using Streamlit, a user-friendly Python library for creating web applications. The deepfake detection model is powered by the facenet_pytorch library for facial recognition and the pytorch_grad_cam library for Grad-CAM visualization. The model itself is based on the InceptionResnetV1 architecture, pretrained on the VGGFace2 dataset.
Challenges we ran into
Developing an effective deepfake detection model comes with its challenges. Some of the key challenges include handling diverse image inputs, optimizing the model for real-time predictions, and ensuring a seamless user experience. Integrating Grad-CAM visualization to enhance transparency and interpretability also posed technical challenges.
Accomplishments that we're proud of
We are proud to have successfully implemented a deepfake detection model and integrated it into an interactive and user-friendly web application. The combination of facial recognition and Grad-CAM visualization allows users to not only receive predictions but also understand the specific features influencing those predictions.
What we learned
Through the development of Cipher-Vision, we gained valuable insights into the complexities of deepfake detection and the importance of interpretability in machine learning models. We improved our skills in working with deep learning libraries, model deployment, and creating engaging user interfaces.
What's next for Cipher-Vision
The future of Cipher-Vision includes expanding its capabilities to encompass video deepfake detection. We are actively working on enhancing the application to analyze and identify manipulated content in videos. By staying at the forefront of deepfake detection technology, we aim to contribute to the ongoing efforts in combating misinformation and raising awareness about the potential risks associated with synthetic media.
Stay tuned for updates as we continue to refine and expand Cipher-Vision to address the evolving landscape of deepfake threats.
Log in or sign up for Devpost to join the conversation.