3D R50, pretrained, pytorch version weights

This is the pretrained weights for I3D, FTCN, and AltFreezing. In their original codes, they use "slowfast" library for loading the pickle file, which can not be directly loaded by pytorch. Therefore, I convert the original pickle file to pytorch version (see the attached .pth file).

The necessary files for Face X-ray (FF++ real + fake)

Including:

1. the face landmarks for each frame. see nearest_face_info.pkl
2. the nearest frame using landmark matching. see landmark_dict_ffall.pkl

Release the pre-trained weights for all detectors

For each detector, we release their best-performing checkpoints on the training set, which can be used for testing and evaluating on other datasets.

Supplementary Material & Pretrained Weights & Dlib Tools

1. Supplementary Material
2. Pretrained Weights
3. Dlib face detector (81 landmarks)