Hybrid Artificial Immune System for Controlled Procedural Modeling

Chengqiao Lin,Ling Fang,Sihan Li,Bin Luo,Juncong Lin
DOI: https://doi.org/10.1109/icpeca53709.2022.9719153
2022-01-01
Abstract:Although powerful in automatic generation of complex geometric structures, procedural modeling is also known to be notoriously difficult in controlling. Various attempts have been made to address the issue in a reverse manner, solving for grammars and parameters of a procedural model to satisfy given target constraints. However, these attempts are suffered from high time cost. In this paper, we present a novel method for controlled procedural modeling, with Artificial Immune System (AIS) as the basis of the framework. By imitating the mechanism of biological immune response, the framework evolvs a set of individual models towards more intuitive high-level goals, such as a volume to be filled with a tree produced procedurally or a silhouette to be matched by a procedurally produced city’s skyline. These goals are intuitive and could be easily specified by artists. Compared with existing solutions, our AIS based framework shows faster convergence than Stochastically-Ordered Sequential Monte Carlo (SOSMC) based method, generates better models than Reversible Jump Markov Chain Monte Carlo (RJMCMC) based work in the long run, and has greater potential in GPU parallelization than Genetic Algorithm (GA) based framework.
What problem does this paper attempt to address?