Can you provide the pretrained model on FF++ (HQ)?

[jiaming-lee]

[jtchen0528]

Hi,

I have trained a few models, still working on how to better generate a more dynamic I2G dataset. The picture showed in README is trained by Deepfake real data, I2G augmented data and fake data, just like the in-dataset setting described in the paper. It gives a rather good result on Deepfake alone. I updated the I2G script to random generated convex hull with by randomly selecting face components. Hope this method can get a better performance with only real images and I2G augmented data.

[twowwj]

I want to test my own data using your method, so, the pre-trained checkpoint is necessary. Could you upload it?