A tool to synthetically generate a Manila-based city. Made using Unreal Engine 4.
Bruce Anakin Miranda, Jacob Manzano, Jan Carlo San Juan, Jaso Jacob Santiago
Manual content generation of complex systems is tedious and time consuming which makes many projects that rely on the production of complex systems infeasible to do and maintain. Procedural content generation (PCG) solves this problem by providing a limitless source of generated content, but will always be only good at generating the complex systems it was designed to generate. Because technology is starting to become more accessible and efficient, countries have the opportunity to utilize and adapt these new technologies for the betterment of society. For example, cities can use computer-vision to create surveillance systems to monitor the flow of traffic, making research for city planning much easier. With simulations of tasks such as computer-vision requiring a large amount of synthetic data, this makes procedural city generation important. There exists software that deals with city generation, with Parish and Müller’s CityEngine being the most influential of the ones publicly available, most software only handles generation of generic cities. There is a lack of study on city generation that is specialized to create cities based on a specific country. This paper proposes a system that is capable of procedurally generating cities that attempts to closely mimic the aesthetics and qualities of Philippine cities.
Follow the instructions below to compile and run the project from source.
- Install Unreal Engine 4.26.2 & its dependencies
- Install Microsoft Visual Studio 2019
- Clone repo (with LFS)
- Download the CONTENTS folder and extract to the root folder
- Right click IECAPSTONE.uproject and click "Generate Visual Studio project files"
- Open the solution file and build
- 8 Core CPU
- 8 GB RAM
- GeForce GTX 10 Series or AMD RX500 Series or above
Other platforms are currently not supported.
Our training procedure is done in batches for each real-life reference image. The script that handles this is located in ./scripts/testing/run.py. To successfully run the test, you need to have a folder within the same directory named images. Inside it are two (2) subfolders: true and unreal. Inside true are your reference images. The images inside need to follow a naming convention and need to be a png. The name of the images need to be m_(n).png where n is sequentially numbered with how many you want to test. For example if you want to test three (3) reference images, there will be m_1.png, m_2.png, and m_3.png. Inside unreal are the Seeds.txt file generated when generating from the application, and two (2) other folders: skewed and unskewed. Inside the skewed folder are the assisted-generation images and inside the unskewed folder are the non-assisted-generation images. Keep in mind that all of these images need to be the same dimension. Once everything is correctly placed, you can then simply execute run.py.
python run.py
We would like to acknowledge De La Salle University (DLSU), Sir Neil Patrick DelGallego, and Dr. Rafael Cabredo.