Image generation method based on stepwise inverse diffusion generative adversarial neural network

The invention discloses an image generation method based on a stepwise inverse diffusion generative adversarial neural network, and the method employs a generative adversarial network GAN to fit the inverse diffusion process of a diffusion model DM, and is used for generating an image more efficiently. The method comprises the following steps: firstly, converting an original image in a data set into Gaussian noise in a mode of gradually adding noise to the image, taking the image subjected to noise addition in each step as a sample to train a generative adversarial network, and learning and generating the image before noise addition by a generator in the GAN through the image subjected to noise addition in each step; the discriminator guides and controls gradual generation of the generator by discriminating a real sample image and the generated image, and finally the generator restores and generates an original image from Gaussian noise through gradual generation. Compared with an image generation method base