Landscape design concept generation combining cultural mapping technology and multimodal modeling

Area of Science:

• Computer Science
• Artificial Intelligence
• Computational Design

Background:

• Traditional landscape design is often inefficient and lacks creativity.
• Existing automated methods struggle with flexibility and balancing structural needs with user preferences.

Purpose of the Study:

• To develop a novel automated landscape design model.
• To enhance creativity, flexibility, and detail in automated landscape generation.
• To integrate multimodal learning for improved design outcomes.

Main Methods:

• Utilized a deep learning model combining Pix2Pix Generative Adversarial Network (GAN) for sketch-to-image generation.
• Employed a CLIP model to align generated images with textual descriptions.
• Implemented sketch-based generation and multimodal learning.

Main Results:

• The proposed model generates higher quality landscape images compared to Pix2Pix and CycleGAN.
• Generated images demonstrate better alignment with textual descriptions.
• Quantitative and qualitative analyses confirm the model's effectiveness in producing innovative and aesthetically pleasing designs.

Conclusions:

• The new model offers a more creative, flexible, and detailed solution for automated landscape design.
• It provides an efficient, innovative, and contextually relevant approach for designers.
• Further development aims to enhance processing of complex design requirements for intelligent design systems.