The Research And Application Of Generating Images And Algorithms Based On The Description Of AttnGAN

Due to its powerful feature learning and feature expression capabilities,GAN has achieved great success in the field of text-generated images.Although GAN has high research value,due to the lack of fine-grained text description at the word level and understanding of the sentence vector,sometimes the generated results deviate greatly from expectations.In order to solve the problem,scholars proposed AttnGAN.Based on the theory of GAN,the model adds Attention mechanism to achieve fine-grained text description to generate images.This paper first introduces the text description generation image model based on AttnGAN.Combining with the scene graph image generation algorithm,the text generation multi-object image algorithm SG-Attn based on AttnGAN model is proposed.Based on a lot of experimental analysis,SG-Attn achieved the goal of generating multi-object images based on text description.The main innovations of this article are as follows:1.The SG-Attn algorithm is proposed.An Attention-generating adversarial network model based on scene graph algorithm(SG-Attn)can automatically obtain the positional relationship between entity,and obtain the scene graph corresponding to the text description through the entity and positional relationship.Using Scene Graphs as an explicit semantic expression tool as an input instead of natural language to describe conditional semantics can allow the model to better learn the characteristics of the positional relationships between objects.Multi-detail features and relatively accurate physical location relationships.2.Develop the SG-Attn system using Qt Designer tool and PyQt tool.The system implements the main function of generating images based on text descriptions.At the same time,the system also provides the function of model training parameter adjustment and visualization of the generated results,enabling users to train and use their own models for text generation images.