无线电通信技术2025，Vol.51Issue(3)：486-492,7.DOI:10.3969/j.issn.1003-3114.2025.03.007
基于多语种文本符号的艺术图像生成模型
Art Image Generation Model Based on Multilingual Text Symbols
摘要
Abstract
Using only text symbols to generate images is called Text-to-Image(TTI)task,which has important application pros-pects in art design.Due to the lack of annotated image data in different languages,TTI research mainly focuses on English,and exist-ing TTI models cannot use existing data in other languages to generate images.Based on the above consideration,the potential of Multi-lingual TTI(MTTI)and current neural machine translation-guided MTTI systems is investigated.By leveraging multilingual text sym-bols,an Art Image Generation Model Based on Multilingual Text Symbols(AIG-MTS)is proposed.By learning weights and integrating knowledge of multilingual texts,differences between languages can be reduced and model performance can be improved.Experiments are conducted on standard datasets COCO-CN,Multi30K Task2 and LAION-5B,and the AIG-MTS model performs best on all datasets when compared with mainstream algorithms.关键词
设计领域/多语种/文本生成图像/多模态编码器/神经机器翻译Key words
design field/multilingual/TTI/multimodal encoder/neural machine translation分类
信息技术与安全科学引用本文复制引用
唐宏,卓诗语..基于多语种文本符号的艺术图像生成模型[J].无线电通信技术,2025,51(3):486-492,7.基金项目
四川省科技计划资助(2022YFG0186) Sichuan Science and Technology Program(2022YFG0186) （2022YFG0186）
