Many pretext tasks lead to representations that are covariant with image transformations. We argue that, instead, semantic representations ought to be invariant ...
確定！ 回上一頁
