The challenge of representation learning: Improved accuracy in deep vision models does not come with better predictions of perceptual similarity

SSLRe-assign community