Automatic Generation of Floorplans using Generative Adversarial Networks
Open Access
Author:
Galiveeti, Sneha Suhitha
Graduate Program:
Computer Science and Engineering
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
June 15, 2021
Committee Members:
Chitaranjan Das, Program Head/Chair Daniel Kifer, Thesis Advisor/Co-Advisor C Lee Giles, Committee Member Sharon Xiaolei Huang, Thesis Advisor/Co-Advisor
Keywords:
GAN Image Synthesis Problem Floorplan Generation
Abstract:
In the present day, demand for construction of houses is increasing rapidly. But creating
and designing a floorplan requires a lot of creativity, technical knowledge and mathematical
skills. The number of architects with these skills are not adequate to meet the
requirements of the rapidly growing demand. We can use Machine Learning to solve this
problem of floorplan generation. This project explores the idea of generation of multiple
floorplans using deep learning models especially Generative Adversarial Networks(GANs).
This work concentrates on the generation of rasterized of floorplans. The main approach
is to let GAN treat floorplans as raster images and learn their distribution to
produce new floorplans. This work explored multiple models of GANs like simple DCGAN,
LSGAN, WGAN, StyleGAN etc. and studied the pros and cons of each model
over two major datasets Structure3D and Graph2Plan. This work also explored the
conditional generation of floorplans i.e., controlling the layout of generated floorplans by
giving input condition to the models in terms of types of rooms present.
