Hi @davide-coccomini,
As you mentioned in the paper, that you have used both FF++ and DFDC dataset to train the model:

• Which version of FF++ dataset do you use (raw/c23/c40)?
• As for DFDC only the training data is available and the public and private test sets are not released. How do you use the DFDC dataset? Do you do a split from the training set itself into train/val/test?