Built using flask and
- MTCNN for face detection (https://github.com/davidsandberg/facenet)
- InceptionResNetV1 + DBSCAN for actor face clustering (https://github.com/davidsandberg/facenet)
- EfficientNet for deepfakedetection per face (https://github.com/lukemelas/EfficientNet-PyTorch)
Clone repository and set up environment by:
$ pip install -r requirements.txt
Once app is up, connect to localhost:5000. From there, paste any YouTube link into the search bar to estimate the deepfakeness of each actor in the clip. Currently, a relatively weak model is deployed, this will be changed in future updates.