ganplan

Using deep learning generative models for analysis of urban morphology - code for MSc dissertation in Smart Cities and Urban Analytics at the Centre for Advanced Spatial Analysis, University College London

The title of this repository is a bit out of sync with its contents, because initially I intended the work to be about using generative adversarial networks for planning support. But eventually I used a modification of variational auto-encoder for analysing city structure. The planning support bit might still get added at some point!

Outline of contents

The repository contains several directories related to different parts of the project:

• tileserver - contains the configuration files and styles for a tileserver-gl setup
• tileclient - contains simple scripts for downloading large sets of tiles from the local tile server
• python - contains python code, which was used for processing the map tiles, using the trained models, and performing the analysis of results
• hpc - contains job submission scripts which were used on UCL servers operating with the Sun Grid Engine batch job system
• R - contains R code, which was used for spatial operations like extracting the urban area boundaries, and for visualising the results of the analysis

Getting started

More information about setting up and running the project will appear here soon. For now, you can browse the python and R code that I used for creating the dataset, running the models on the UCL high performance computing nodes, perform an analysis of the results, and visualise the results.