This AI designs beautiful Forest Landscapes for Games!
Overview of the paper “Data-driven Authoring of Large-scale Ecosystems” by K Kapp et al.
Designing large areas of forests in games can usually be a time consuming task for game designers. Randomly placing trees and plants of any species around complex terrains like hills and valleys do not yield very realistic results, so knowledge about vegetation is also required alongside a lot of manual efforts to design a good forest. Thankfully, this is also a task where Machine Learning can be used to let the computer design realistic forests by learning from real-world data.
Our paper in focus today does just that. It is titled “Data-driven Authoring of Large-scale Ecosystems” and is published at this year’s SIGGRAPH conference in Computer Graphics.
It provides an interactive tool where you can easily define the style of the forest you want to build in your game, and the rest is taken care of by the Deep Learning models in the background. The final output can be rendered immediately so you can quickly assess the output and make any necessary changes, thereby saving a whole lot of time and efforts. This Deep Learning method learns about different types of plants and vegetation like canopy height, under-storey layer and forest floor.
Conditional GANs are used to generate this information, where the output is conditioned on inputs like terrain elevation levels and optionally also forest density map for added user control. This makes it ideal for designing games where we want at least a little bit of control over the GAN’s generated output but without requiring too much time or effort.
As you can see, the results presented in the paper look fabulous. Even though you may not notice this when playing the game, this type of work dramatically reduces the development time of such games with large open-world areas with the help of deep learning. If you are a designer, I would highly recommend you check out such Deep Learning aided design tools as they will make your life much easier and your work more productive.
Useful Links
Thank you for reading. If you liked this article, you may follow more of my work on Medium, GitHub, or subscribe to my YouTube channel.
