In the ever-evolving world of computer rendering technologies, the demand for more original and diverse 3D models is ever-increasing to the point where the man-hours for creating these models for video games, movies and architectural designs are counted in hundreds, if not thousands. This project proposes a new way to generate relatively large voxel models from smaller example voxel models given by the user. Basing on the Model Synthesis concept this project proposes two different models that seeks to use the volumetric data found in the voxel models to extract the patterns found in the voxel model given by the user and combine those patterns in new and unexpected ways all the while staying coherent to the input model structure without the use of any rules or grammars. While the first results are small scale, they show promise of real application with further optimisations. The framework written with this paper aims to lay the foundation for the future developments of model synthesis for voxel models.
keywords: Procedural Modeling, Texture Synthesis, C#, IC, Thesis Project, Machine Learning, Masters, Unity3D, Voxel, Convolutions
master’s thesis project student: Matvey Khokhlov
thesis supervisor: Immanuel Koh
thesis professor: Jeffrey Huang
invited external guest: Paul C. Merrell (Google)
Image credits: Immanuel Koh