Visual design and digital transformation of imaging under artificial intelligence technology

Area of Science:

• Artificial Intelligence
• Computer Vision
• Human-Computer Interaction

Background:

• The design field is undergoing digital transformation, necessitating advanced tools for human-machine collaboration.
• Current AI models often struggle with integrating diverse design inputs and maintaining control over generated outputs.

Purpose of the Study:

• To propose a novel conditional hierarchical generative adversarial network (GAN) for intelligent visual design.
• To enhance human-machine collaboration and facilitate digital transformation in the design industry.

Main Methods:

• Developed a conditional hierarchical GAN with a multimodal conditional fusion encoder, hierarchical generator, and collaborative multi-discriminator.
• Integrated multiple design intents (text, sketches, layouts, brand guidelines) using a cross-modal attention mechanism.
• Employed a two-stage "structure first, details later" generation strategy and multi-objective adversarial training.

Main Results:

• Achieved a Fréchet Inception Distance (FID) score of 18.37 on the COCO-Stuff 164K dataset, indicating high generation quality.
• Reached 92.8% layout control accuracy, demonstrating superior controllability compared to baseline models.
• Validated the model's effectiveness in professional visual design tasks.

Conclusions:

• The proposed conditional hierarchical GAN offers an innovative technical solution for intelligent visual design.
• The study provides insights into AI-driven creative agency, human-machine collaboration, and digital transformation in the AI era.