Нормалізація зображень за допомогою генеративних змагальних мереж

Abstract

The problem of recognition and reconstruction of damaged or low-quality images was solved. The problem of generating potential prototype or corrected images was solved. The algorithm of generative adversarial networks functioning was discussed, researched and described. Potential use cases and algorithm purposes in the field of image normalization were proposed. In practice, dataset serves the initial training data for the discriminator. Training the discriminator involves presenting it with samples from dataset, until it reaches certain level of accuracy. Typically, the generator is seeded with a randomized input that is sampled from a predefined latent space. Backpropagation is applied in both networks so that the generator produces better images, while the discriminator becomes more skilled at flagging synthetic images.