1 paper · 1 filter
Steffen Czolbe, Oswin Krause, Ingemar Cox +1
To train Variational Autoencoders (VAEs) to generate realistic imagery requires a loss function that reflects human perception of image similarity. We propose such a loss function…