Data Driven Procedural Generation

This repository presents some frontier researches using data driven methods, e.g. machine learning, for procedural content generation purpose. Literally, "procedural" means rule-based, the purpose of machine learning is to learn rules from data. This is paradigm-shifting since tech-artists are used to creating rules to generate assets, but their roles will transform into supervising and guiding computer to understand rules with data-driven methods.

Some Sources

https://github.com/chenweikai/3D-Machine-Learning
http://www.creativeai.net/

To Contribute

Please add through pull requests or open issue.

Environment

Terrain

A step towards procedural terrain generation with GANs paper code

StreetGAN: Towards Road Network Synthesis with Generative Adversarial Networks paper

Building

FrankenGAN: Guided Detail Synthesis for Building Mass-Models Using Style-Synchonized GANs paper code

Learning Design Patterns with Bayesian Grammar Induction paper

Guided proceduralization: Optimizing geometry processing and grammar extraction for architectural models paper

Bayesian grammar learning for inverse procedural modeling paper

Interactive Design of Probability Density Functions for Shape Grammars paper

Level Design

Adversarially Tuned Scene Generation paper

Deep Convolutional Priors for Indoor Scene Synthesis paper

Make It Home: Automatic Optimization of Furniture Arrangement paper

Game Level Generation Using Neural Networks news

Character

Hair

3D hair synthesis using volumetric variational autoencoders paper

HairNet: Single-View Hair Reconstruction using Convolutional Neural Networks paper

Texture

Stylizing Face Images via Multiple Exemplars paper

Photorealistic Facial Texture Inference Using Deep Neural Networks paper