Score-Guided Generative Adversarial Networks

We propose a generative adversarial network (GAN) that introduces an evaluator module using pretrained networks. The proposed model, called score-guided GAN (ScoreGAN), is trained evaluation metric for GANs, i.e., the Inception score, as rough guide training of generator. Using another instead network, ScoreGAN circumvents overfitting such generated samples do not correspond to examples network. In addition, metrics are employed only in auxiliary role prevent overfitting. When evaluated CIFAR-10 dataset, achieved score 10.36 ± 0.15, which corresponds state-of-the-art performance. To generalize effectiveness ScoreGAN, model was further CIFAR-100. outperformed other existing methods, achieving Fréchet distance (FID) 13.98.