Image generation model training method and device, equipment and computer storage medium

Abstract

Embodiments of the present application provide a training method, device and equipment of an image generation model and a computer storage medium. The training method comprises: obtaining a collocation scene graph, the collocation scene graph comprising at least one object, each of the at least one object corresponding to an object main body graph, the object main body graph comprising rendering information of the object; performing light and shadow removal processing on the collocation scene graph by using an albedo estimation algorithm to determine a light and shadow removed graph; taking the collocation scene graph as a true value, and supervising training of a diffusion model based on the light and shadow removed graph and the object main body graph corresponding to each of the at least one object to obtain an image generation model, the image generation model being trained to perform image rendering operation based on the object main body graph to obtain a rendered scene graph. In the embodiments, when the image generation model obtained by training is used to perform image generation operation, not only the reality of image rendering is ensured, but also the computing resources required for image rendering operation are reduced.

Images