Scholarlink

<p>To solve the three core problems existing in the current GUI automatic generation technology, namely unstable generation quality, inconsistent style and unreasonable layout, a GUI-assisted generation model based on generative adversarial network (GUI-GAN) is proposed in this study. The research method achieves end-to-end GUI generation by constructing a multi-objective dynamic optimization framework, integrating the LSTM sequence generator and CNN discriminator, and introducing style contrast loss and structured representation mechanisms. The research results showed that GUI-GAN significantly outperformed the existing methods in terms of generation quality. The FID value dropped to 72, the cosine similarity of the style vector reached 0.93, the component alignment accuracy remained at 88.0% in complex scenarios, and the layout density deviation was controlled within 0.06. Practical application tests showed that its development efficiency score reached 4.52, user satisfaction score was 4.28, and functional integrity score was 4.60. The conclusion shows that this model achieves high-quality and highly consistent GUI automatic generation by explicitly optimizing style consistency and layout rationality, providing a deployable solution for intelligent design tools. This research is significant because it is the first to model style consistency as a quantifiable optimization objective. This promotes a paradigm shift in GUI generation, moving from local optimization to multi-objective collaborative control.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Construction of auxiliary generative models for GUI design based on generative adversarial networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月27日

查看全文
  • Xiaoxia Li

摘要

To solve the three core problems existing in the current GUI automatic generation technology, namely unstable generation quality, inconsistent style and unreasonable layout, a GUI-assisted generation model based on generative adversarial network (GUI-GAN) is proposed in this study. The research method achieves end-to-end GUI generation by constructing a multi-objective dynamic optimization framework, integrating the LSTM sequence generator and CNN discriminator, and introducing style contrast loss and structured representation mechanisms. The research results showed that GUI-GAN significantly outperformed the existing methods in terms of generation quality. The FID value dropped to 72, the cosine similarity of the style vector reached 0.93, the component alignment accuracy remained at 88.0% in complex scenarios, and the layout density deviation was controlled within 0.06. Practical application tests showed that its development efficiency score reached 4.52, user satisfaction score was 4.28, and functional integrity score was 4.60. The conclusion shows that this model achieves high-quality and highly consistent GUI automatic generation by explicitly optimizing style consistency and layout rationality, providing a deployable solution for intelligent design tools. This research is significant because it is the first to model style consistency as a quantifiable optimization objective. This promotes a paradigm shift in GUI generation, moving from local optimization to multi-objective collaborative control.