Facade Style Mixing using Artificial Intelligence for Urban Infill

Artificial Intelligence and especially machine learning have noticed rapid advancement on 1 image processing operations. However, its involvement in the architectural design is still in its initial 2 stages compared to other disciplines. Therefore, this paper addresses the issues of developing an 3 integrated bottom up digital design approach and details a research framework for the incorporation 4 of Deep Convolutional Generative Adversarial Network (GAN) for early stage design exploration and 5 generation of intricate and complex alternative facade designs for urban infill. This paper proposes a 6 novel building facade design by merging two neighboring building’s architecture style, size, scale, 7 openings, as reference to create a new building design in the same neighborhood for urban infill. 8 This newly produced building contains the outline, style and shape of the parent buildings. A 2D 9 urban infill building design is generated as a picture where 1) neighboring buildings are imported 10 as a reference using mobile phone and 2)iFACADE decode their spatial adjacency. It is depicted the 11 iFACADE will be useful for designers in the early design stage to generate new façades depending 12 on existing buildings in a short time that will save time and energy. Besides, building owners can use 13 iFACADE to show their architects their preferred architecture facade by mixing two building styles 14 and generating a new building. Therefore, it is depicted that iFACADE can become a communication 15 platform in the early design stages between architects and owners. Initial results properly define a 16 heuristic function for generating abstract design facade elements and sufficiently illustrate the desired 17 functionality of our developed prototype. 18


21
In general, building envelope design configurations follow similar urban street configurations  what it has seen in the past; the same process that we humans undergo when we dream. As every neighborhood has its own regulations for a modular architecture style that all buildings 104 within the neighborhood should follow. Therefore most buildings share the same general building 105 characteristics, colors, window openings, and style. Therefore iFACADE was built to generate a new 106 building elevation depending on existing neighborhood building styles.

107
The iFACADE framework is divided into four essential stages: a) input reference neighborhood 108 buildings where it needs two reference building facade 2d colored images (B1 and B2) as input and 109 neighborhood architecture policies as resource test to help the architect evaluate generated facade model and upgraded its system architecture depending on the architect and user's feedback.

116
The system is built in a cloud and front-back end web site is developed to host trained models. It 117 is critical to have the website to let the users interact easily from anywhere by logging to it. Currently it 118 only has one functionality which is generating style mix images and it could be extended to host more 119 information about each neighborhood's regulations in the future. The user submits a review report 120 and it helps the iFACADE team to evaluate the iFACADe model, upgrade its system architecture and 121 retrain it. Besides, uploaded images by users will be added to the training dataset and expand it which 122 will aid the iFACADE team to generate better and more accurate facades in the future as shown in

129
In General, iFACADE model consists of a generator that creates a fake facade image depending 130 on real image input and a discriminator that determines whether the input image is fake or real.
Where x i is a normalized instance that we apply AdalN to, y is a set of two scalars (y s , y b ) that 141 control the "style" of the generated image, f (w) represents a learning affine transformation.

142
In the proposed generator, AdalN operation is normalized to map networks that show various where p(z) is the standard Gaussian distribution. The generator's conditional mapping function

152
The discriminator is an unsupervised classifier that determines if the input image is true or fake.

153
The model architecture of discriminator used in iFACADE is adopted by projection discriminator 154 proposed by [6]. Differently from the approach of concatenating the embedded conditional vectors 155 into feature vectors, the projection discriminator integrates conditional vectors.  is generated by our model given different random noises. We can see that random noise plays a role 241 in expressing differences such as food topping. Figure-4 shows that images generated by our model 242 simultaneously with condition vectors representing two or more classes. We can see that our model 243 can generate each feature even when given the multiple condition vectors.

Style transfer
where G is the generator, and is a vector whose components are small numbers that are sampled 245 randomly.

246
The iFACADE does not work on neighborhoods that do not follow modular architecture styles as 247 generated facades will have characteristics from references building facades. It also cannot apply on images. Finally, we hope that our proposed model will contribute to further architecture design studies.

268
Besides, iFACADE will be extended to generate 3d facade elements beside the 2d images for it to be 269 used directly in the architecture early design stages directly and increase automation.