Abstract
For almost a century, cities have developed and grown in scale partly to manage and realize the advancements of technology, transforming and evolving both our natural and societal environment. However, as cities grow in size, they become increasingly hard to manage and understand. Our current methods of planning: a mixture of top-down planning and commercially incentivized project development has its limits in addressing the challenges facing for our society as a whole; rendering our growth and developed unsustainable. Other than economic and environmental pressures, the aesthetics of many cities are becoming diluted by dry and ubiquitous planning codes and profit-driven designs.
To realize the design required for today’s world, we need a deeper understanding of the morphology of cities in order to respond to the challenges facing urban development. We need the ability to utilize data and new technologies in a more creative and functional way.
The goal of this study is to envisage the future of Xiongan through pattern design and Deep Neural Style Transfer, aiming to create adaptable large scale planning solutions. This study demonstrates how new technologies can provide us insight into the various possibilities of a given site and seek out to understand how Style Transfers can affect and inform the designer.
The area undergoing the planning process as part of the new Xiong An area spans across an area more than 20 km from East to West and more than 15 km in the North-South direction. This map is at an elevation of 20,000 meters.
This image shows the elements used and how they can be combined into more complicated designs.
The morphology process is addressed by using patterns in scale. In stage 1, the center is established, branches are laid out for future growth. In stages 2 and 3, the center is reinforced by the establishment of smaller urban centers around its periphery. In the final stage, satellite centers are reinforced as well, suggesting a maturation of its growth process.
The first two columns, one in black and white, and the other with color, shows how different elements and patterns are layered. The third column shows how a specific pattern is laid out, styled and texturized.
The patterned base map is the result of many elements and patterns woven together, and its base map for the style transfer process. These elements are superimposed, though arbitrary, it is highly flexible and suggests limitless possibilities.
The Urban texture StyleGAN 1 is generated by using an aerial photograph of Barcelona, its texture is transferred onto an element created for this project. To capture and transfer the texture at the right scale, the aerial photo of Barcelona is taken at an elevation of 20,000 meters, same as the Xiong An aerial map. Barcelona is used because it is well organized and pedestrian friendly. Other than its urban features, it has a population density of about 10,000 people/km2 , similar to the planning goals of Xiong An.
Detailed view of Urban StyleGAN 1
Detailed view of Urban StyleGAN 2. This StyleGAN used a different pattern and a different city. For this StyleGAN, the aerial photo used is from Venice, at an elevation of 5000m. This StyleGAN is used to demonstrate how different scales can be applied for different purposes.
Using part of the Xiong An map, from an elevation of 5000m. A direct style transferred from pattern to base generates the Stylized map.
By transferring Urban StyleGAN 2 onto the Stylized map, unique urban features are created in the process.
This image explains the work flow on how different Stylized maps used for this project is created. The elements on the left are used for two different purposes. First, elements are combined to generate the Patterned Xiong An map. Second, these elements are organized into different Styles, these Styles are intended to reflect different planning solutions. In the end, variations of the Patterned base map are created via Deep Neural Networks, where the Styles are transferred onto the Patterned base map.
Here the results of transferring Urban StyleGAN 1 onto the various styled patterns are shown.
StyleGAN 1 is the first of three Styles created for this project. This Style is composed of a center pattern layered on top of a light-colored background.
This map is the result of transferring StyleGAN 1 on to the Patterned base map. The result captured most of the white background, while other parts of the pattern are emphasized.
This map is the result of transferring StyleGAN 1 with a higher resolution.
After its initial style transfer, Urban StyleGAN1 is used to apply urban texture onto the stylized map.
StyleGAN 2 is the second of three Styles created for this project. This Style is composed of various centers and densely woven together on top of a dark-colored background.
This map is the result of transferring StyleGAN 2 on to the Patterned base map. By using a denser composition of patterns, the result is colorful and the contrast is strong.
This map is the result of transferring StyleGAN 2 with a higher resolution.
After its initial style transfer, Urban StyleGAN1 is used to apply urban texture onto the stylized map.
StyleGAN 3 is the last of three Styles created for this project. This Style is composed of various centers but loosely posed together on top of a dark-colored background.
This map is the result of transferring StyleGAN 3 on to the Patterned base map. By using a looser composition of patterns, the result drastically different from a denser application. The colors in this image is concentrated and group, each forming its own cell.
This map is the result of transferring StyleGAN 3 with a higher resolution.
After its initial style transfer, Urban StyleGAN1 is used to apply urban texture onto the stylized map.
This map shows the Deep Neural Style Transfer process, by demonstrating how the original plot of land can first be planed, stylized and texturized through different designs.
A{AI} 