WANQI SHI
Finding the Implications of the COVID-19 on Dwelling Plans with Deep Learning: A Shanghai Case Study
Summary
This research aims to explore the implications of COVID-19 on future dwelling design and Design Standards for Residential Buildings in Shanghai. The implication, including physical and psychological requirements, relates to changed residential requirements and future house preferences for dwellings. Through interviews and surveys on lived experience during the pandemic and the DL model, the gaps between the expected and current dwellings caused by changed residential requirements will make the quantitative conclusions and advice on the future direction of optimizing dwelling more objective and understandable.
Additional info
Research Question
In this context, the main research questions relates to people's lived experiences and domestic use during the lockdowns. This research, at the same time, due to the characteristics of Shanghai residential design, would involve the Design Standard for Residential Buildings, the mandatory document as part of the research.
How has COVID-19 affected the use and experience of dwellings?
How could people's spatial perception, as affected by the Covid-19 pandemic, alter the plans of dwelling?
To what extent can this new quantifiable data set influence residential design standards in the longer term?
Methodology
Firstly, in-depth surveys and interviews were conducted to understand residents’ lived experiences during the pandemic and the possible optimized direction for future dwelling plans. Through interviews, respondents would make optimizations on their dwelling layouts and plans combined with their changed residential requirements brought by the pandemic, which is beneficial for transferring abstract house preferences into architectural plans. Secondly, I trained a Deep-Learning model with 90 pairs of actual and colored dwellings plans. The trained DL model based on PIX2PIX (Image-to-image translation with conditional adversarial networks) can automatically identify and distinguish each functional area of the plan with color blocks and could meanwhile generate actual dwelling plans based on the color blocks. Thus, I input the actual and optimized plans from respondents and got the colored dwelling plans identified by different function. Thirdly, using the computer vision code OpenCV, I obtained the proportion of each function area on the basis of colored dwelling plans. Thus, made comparisons of these data in order to acquire quantitative conclusions.
Discussion
The outbreak of COVID-19 was more like an opportunity for change, for people to realize the intense connection they had never had with the residential space. The significance of my research lies in its potential to address the unequal distribution of housing conditions in the long term, as the restrictive document for optimizing housing is to raise the minimum housing conditions. Even small changes in spatial conditions will have a certain impact on basic dwellings such as low-income residences. Meanwhile, by collecting quantitative and qualitative data, more substantial evidence can be used to revise and improve the Design Standard for Residential Buildings and guide future dwelling design.
Bio
Wanqi is an active researcher and a Mres candidate at the Royal College of Art. She is interested in the field of the application of AI models in architecture and urban studies. During her undergraduate years, she worked as a research assistant on a project in Suzhou's underground space. Later, she also participated in the study with Dr. Kew from Cornell on the Beijing AI carbon emission project. During the postgraduate period, I also participated in the workshop research on Architectural Reinforcement learning hosted by ETH.
Wanqi is also a design architect and worked as an assistant architect on projects such as Shili Lianjiang and Guanyin Pavilion, which are currently constructing in China.
Methodolgy
The Basic Information of 15 Selected Research Subjects
I divided my interviewees into three groups based on their dwelling construction year and the published time of the design code. The construction years of the three groups are from 2001 to 2007, from 2007 to 2013, and from 2013 to 2019, which were all built before the pandemic.
Requirements for Dwellings
15 In-depth interviews covered the field of details of lived experience and residential requirements related to their dwelling. The interviews were recorded, and participants were invited to optimize their dwelling plans in their house preferences. On the basis of interviews, the relationship between requirements and residential attributes, and the upgraded house preferences were acquired.
The Comparison of Original and Optimized Dwelling Plans
Participants used the corresponding color where they want to optimize, which might be to replace the function of the room, or to expand the size of the room.
PIX2PIX to Identify the Function of Dwelling Plans
I set up my training groups A and B. Group A is a dwelling plan colored according to functional areas drawn by myself. Group B is a standardized dwelling plan from Lianjia website.The first training is to identify the dwelling plan from the color block, and the second training is to generate the color block from the dwelling plan
Quantified Analysis
The generation of the plan of dwellings with OpenCV. Each color represents a different functional area. The code identifies the proportion of each color, so the area size of each functional area can be identified through the code. In the first row, 41.3% is the proportion of the bedroom in the picture, and the actual size of the living room can be obtained by calculation.
Quantified Optimization Direction of Future Dwellings
Through the interview of living experience, people's changing needs are mainly reflected in storage space, bathroom, bedroom, living room and indoor and outdoor transition space. Using the years as the dividing line shows the change in the mean, and also shows the gap between the existing homes and people's ideal homes after the pandemic.
Future Dwelling Plan
The PIX2PIX generated the actual dwelling plans based on the optimization and it is able to provide more vital evidence to make internal space standards mandatory and dwelling design guidelines.
