PlanGCN
CS109B Final Project, Spring 2020
Instructor: Pavlos Protopapas (Harvard SEAS), Mark Glickman (Harvard SEAS), Jose Luis García del Castillo y López (Harvard GSD), Chris Tanner (Harvard SEAS), Javier Zazo (Harvard SEAS, TF)
Collaborator: Ao Li, Runjia Tian, Xiaoshi Wang
Location: Cambridge, US
This project seeks to automatically generate residential spatial layout by inputting a room relation graph. We implemented a Graph Convolutional Network to study the relationship between different room types and their spatial connection types.
Rooms in a residential floor plan are categorized into 10 major room types: Living Room, Bedroom, Kitchen, Dining, Bath, Storage, Entry, Garage, Outdoor and Other. Connections are defined as ADJACENT, DOOR-CONNECT, NOT-CONNECT.
We developed a Python library to parse 3,500 structured floor plan SVG files from CubiCasa5K and construct their room relation graphs. The source code and parallel graph dataset, GubiGraph5K, is published on CAADRIA 2021.
GitHub repo: github.com/luyueheng/CubiGraph5K
Live demo available: gcn.luyueheng.com
GCN demo:
Paper presentation for CAADRIA 2021:
Comments disabled