Research On Automatic Generation Of Building Layout Based On Pix2pix

With the popularization of higher education and the development of economy in China entering a new stage of competitive development,the role of higher education in today’s social and economic development is increasingly prominent,and the construction of university is in urgent need.Campus layout is an important stage in the early stage of campus program,which determines the spatial organization of campus and the development direction of the overall program.However,due to the large scale and complex functions of the campus,layout design is largely limited by the architect’s personal professional knowledge and experience.In recent years,artificial intelligence technology has been booming,and deep learning technology has shown the potential of artificial intelligence to imitate human thinking.With the proposal of pix2pix and other neural networks,the ability of deep learning in image processing has attracted the attention of the scholars of architecture.At the same time,cross-disciplinary research on the combination of artificial intelligence and architectural design has gradually emerged.Taking the layout of university campus as an example,this paper tries to use pix2pix to realize the automatic generation design of building layout.After the training,the model can quickly generate the required layout sketch of the campus according to the site conditions and surrounding environment in a very short period of time,inspiring architects to carry out diversified design thinking and providing candidate schemes.The two key components of deep learning technology are the algorithm framework and data.According to the technical principle,data used in some deep learning technologies need to be annotated.Similar to a teacher emphasizing the main points of an exam,data annotation is meant to help the computer learn the target information efficiently and accurately.However,different from the simple labeling of human faces in tasks such as face recognition,the data annotation in architectural layout design is the complex labeling that requires professional background.In the crossover application of pix2pix,the algorithm framework is established by scholars in the field of computer science,while the main work of scholars in the field of architecture is to provide high-quality data based on professional knowledge.It can be said that the method and content of data annotation determine to a large extent the building layout effect finally generated by the pix2pix model.This paper is mainly divided into three parts:In the first part of this paper,through sorting out the research and literature of artificial intelligence technology in the field of building layout generation,combining with the existing experimental work,the theoretical basis of the experiment is established,and the experimental difficulties and preliminary solutions are proposed.The second part of this paper is based on pix2pix,taking the central loop linear layout as an example,the experimental process of generating the layout of the university campus is detailed.This part first clarifies the experimental ideas and collects excellent practice cases on university campuses.Secondly,based on the theory of campus layout design,data annotation rules were designed,and a limited number of university campus practice cases were manually redrawn to establish a small data set of ‘small but efficient’,so as to solve the contradiction of ‘large demand for data with few effective cases’.At the same time,through repeated experiments,data annotation rules are optimized,in an attempt to express the core content of architectural design--layout methods and rules--in image data more clearly,and provide them to pix2pix model for learning,so as to achieve the desired layout generation design effect.The third part of this paper is the review and discussion of the whole research.Firstly,the final experimental results were analyzed qualitatively and quantitatively.On this basis,the limitations and shortcomings of the experiment were reflected.Secondly,the strategy of building layout generation based on small data set is summarized.Finally,the layout result generated by pix2pix is compared with the project completed by human architects,and the challenges and opportunities of architects under the tide of artificial intelligence are considered.