Conference Paper/Proceeding/Abstract 123 views 3 downloads
Spatial Layout Generation via Generative Adversarial Networks
International Conference on AI-generated Content
Swansea University Authors:
YUE YANG, Hans Ren, Xianghua Xie
-
PDF | Accepted Manuscript
Author accepted manuscript document released under the terms of a Creative Commons CC-BY licence using the Swansea University Research Publications Policy (rights retention).
Download (2.83MB)
Abstract
The process of architectural floor plan generation is a complex task traditionally performed by architects, requiring a deep understanding of spatial relationships, structural constraints, and aesthetic principles. In recent years, advances in computational design and Artificial Intelligence (AI) ha...
| Published in: | International Conference on AI-generated Content |
|---|---|
| Published: |
2024
|
| URI: | https://cronfa.swan.ac.uk/Record/cronfa68386 |
| Abstract: |
The process of architectural floor plan generation is a complex task traditionally performed by architects, requiring a deep understanding of spatial relationships, structural constraints, and aesthetic principles. In recent years, advances in computational design and Artificial Intelligence (AI) have enabled the automation of floor plan generation, significantly enhancing efficiency and creativity in the architectural workflow. In this paper, we explored the integration of traditional architectural design methods with advanced technology, focusing on the transformative role of Generative Adversarial Networks (GAN) in floor plan generation. In this work, we created a new dataset containing more than 1200 carefully processed images for the automatic generation of floor plans. These samples come from different platforms and are processed to become algorithm-friendly types. We will make the dataset public available. The algorithm we used is the pix2pix network, which is enhanced with a self-attention mechanism for better spatial understanding and spectral normalization for improved output quality. We demonstrate the versatility of the GAN model in generating complex floor plans for various architectural needs based on our dataset. It also addresses challenges such as model stability, detail refinement, and generating non-standard room shapes, offering insights for future advancements in the field. |
|---|---|
| College: |
Faculty of Science and Engineering |