In our experiments on 400 million image-text pairs, FLIP improves both accuracy and speed over the no-masking baseline.
確定! 回上一頁