Adaptive image generation system, method and computer readable medium integrating domain knowledge and graph retrieval-augmented generation mechanism

Abstract

The invention discloses an adaptive image generation system, method and computer readable medium integrating domain knowledge and graph retrieval-augmented generation mechanism. A domain knowledge modeling preprocessing module obtains or inputs graphic and text data of a professional field or domain knowledge from an authorized graphic and text database to construct a multimodal vector database, and then uses the multimodal vector database as a knowledge retrieval source for a multimodal graph retrieval-augmented generation conditional adaptive module. Furthermore, the multimodal graph retrieval-augmented generation conditional adaptive module adaptively specifies a condition required for an image generation target according to a prompt words to compensate for an image generation diffusion model's understanding of unknown concept, and then uses an image generation input condition as input of the image generation diffusion model, in order to use an image generation workflow template selected from an image generation workflow template library to generate an image or image content that match the prompt words.

Images