Creating 3D digital nature is a time-consuming and difficult process that requires a large group of well-trained 3D artists plus expensive 3D models and assets. Especially the layout process is very time consuming.
Imagine you could take any 2D photo of a landscape and automatically convert it into a 3D interactive environment for use in a video game or film production.
Creating intelligent software requires a lot of training. Neural networks need a lot of data (thousands of images) which all have to be labelled, so they can ´understand´ their tasks. Therefore, we have divided the whole project into four main AI modules:
The MOODS module is currently in development and the first prototype is working already! This prototype serves as a proof-of-concept. The software can read any landscape photo and transfer the extracted parameters to the Unreal Engine 4.
The second module is important as without the camera perspective and position, there won´t be any 3-dimensional information which is essential to re-create a 3D scene. The goal is to interpret the most important camera parameters from a single 2D image.
Knowing which objects are visible in an image will help to set up a set of rules to distribute such objects as 3D models within a 3D scene. Together with the camera information, a 3D scene can be recreated using a set of pre-made 3D models and assets.
A tree is a tree is a tree ? Not really, since a tree can have different materials depending on the time of the year. It may look different in winter, spring, autumn or summer. A lake can also have waves, be calm or frozen, depending on the weather and time of year.
The concept of the WUNDERPARC core software is quite simple. An input image is analyzed by a series of neural networks that extract parameters for the 3D scene reconstruction.
2-dimensional image of a landscape
A 2D image of a landscape, any landscape, is the source of the scene you want to create. The source image can be a photo, concept art, matte painting or anything else.
The neural network extracts information
Our neural network is specifically trained to read and interpret information from the source image, which will then be used to define parameters that are needed for a 3D scene.
Parameter set for 3D reconstruction
The extracted parameters include camera position and perspective, which objects are visible, clouds, sky, background, light and mood. This parameter set is the base for the 3D scene creation.
The recreated 3D scene
With the parameters, the scene reconstruction takes place. Using the parameters together with a 3D model library, materials, and procedural systems, the scene is built.
FIRST PROTOTYPE
The first prototype of the MODULE 1 – MOODS reads fog density, color temperature, and light. The parameters are then transferred to the UNREAL ENGINE 4 where the procedural systems are used to recreate a similar mood.
On this blog we will post about new developments, our progress, events and anything related to this project.
Interview lesen (Deutsch) Read Interview (English) After graduating successfully with the MEDIA LIFT Incubator, we just recently were
From August 20. to August 22. we will exhibit at the GameCity Hamburg booth in hall 4.1 in the business area.
We´re very happy that we are approaching a first milestone: The first MOODS prototype which is supposed to be a proof-of-concept for our
SUPPORTED BY:
We´re at the beginning of a new era.
3D content creation is undergoing a major change in the next few years through assisted machine learning software. WUNDERPARC is part of this journey by developing a software that helps anyone to create 3D scenes more easily.
Please get in touch if you´re interested in more…