Instance-Conditioned GAN

Finally, we extend IC-GAN to the class-conditional case and show semantically controllable generation and competitive quantitative results on ImageNet; while improving over BigGAN on ImageNet-L T. Code and trained models to reproduce the reported results are