search

frederickszk / LRNet

Landmark Recurrent Network: An efficient and robust framework for Deepfakes detection
MIT License
90 stars 13 forks source link

questions regarding training #5

Open seanko29 opened 2 years ago

seanko29 commented 2 years ago

Hello,

Why is ur dataset composed of txt files not image files? Can you use image files (ex. FF++ c40 DF,F2F..) as an input instead of txt files?

frederickszk commented 2 years ago

Hello, Our model is not trained in an end-to-end fashion. It firstly extracts landmarks sequence from image files, which generates the txt files. And the subsequent training and inference are only performed on these txt files. Therefore we also provide them for convenience. Besides, you can also use your own landmark detector to generate the landmarks as input. For more details, you can refer to readme.md and our paper. Thanks 😄!