fsfm-3c_vit_base_patch16 pre-trained on FF++_o (~43W facial image from train and val videos of the original/YouTube subset in FaceForensics++ c23/HQ version, extract all frames) for 400 epochs.
fsfm-3c_vit_base_patch16 pre-trained on FF++_o (~43W facial image from train and val videos of the original/YouTube subset in FaceForensics++ c23/HQ version, extract all frames) for 400 epochs.