Fake it till you make it: Learning transferable representations from synthetic ImageNet clones

Open in new window