Thank you for sharing your great job! I trained on my own preprocessed FF++ dataset, but the results are unsatisfactory. Both c23 and c40 are about 3-4 precentage points lower than the valued provided in the paper. Can you sharing some tricks during training process.
Thank you for sharing your great job! I trained on my own preprocessed FF++ dataset, but the results are unsatisfactory. Both c23 and c40 are about 3-4 precentage points lower than the valued provided in the paper. Can you sharing some tricks during training process.