In recent years, generative adversarial nets ( GANs) have developed rapidly and have become one of the main research directions in the current machine learning field. GAN is derived from the idea of zero-sum game. Its generator and discriminator are opposed to learning. The purpose is to obtain the data distribution of a given sample and generate new sample data. A large number of investigations have been made on GAN models in image generation, abnormal sample detection and location, text generation pictures and picture super-resolution. The substantial progress made in the application of these GANs has been systematically explained. The background and research significance, theoretical model and improved structure of GAN, and its main application fields are summarized. The shortcomings of GAN and its future development direction were summarized.