MIT researchers claim augmentation technique can train GANs with less data

MIT researchers propose a new technique -- DiffAugment -- that ostensibly enables the training of state-of-the-art GANs with few images.