BigGAN

BigGAN is a generative adversarial network approach that enables the routine generation of high-resolution and high-quality images by scaling up the batch size and number of model parameters. It addresses the limitations of traditional GANs, which are typically restricted to generating small images and have a fragile training process. BigGAN combines recent best practices in training class-conditional images and is capable of producing large and detailed synthetic images.