SF-GAN: Semantic fusion generative adversarial networks for text-to-image synthesis

Open Webpage

Bing Yang, Xueqin Xiang, Wanzeng Kong, Jianhai Zhang, Jinliang Yao

Published: 01 Mar 2025, Last Modified: 04 Nov 2025Expert Systems with ApplicationsEveryoneRevisionsCC BY-SA 4.0
Abstract: Highlights•A recurrent semantic fusion network to ensure a coherent fusion of text–visual cues.•A contrastive loss to strength the underlying semantics of text and the image.•A dynamic convolution to enable the generator to dynamically produce an image.•A word-level discriminator to capture relationship between word and image subregion.•Experimental results show the efficacy of SF-GAN on the CUB and COCO datasets.