GAN Cocktail: mixing GANs without dataset access

Omri Avrahami, Dani Lischinski, Ohad Fried

Today's generative models are capable of synthesizing high-fidelity images, but each model specializes on a specific target domain. This raises the need for model merging: combinin…