Conceptos Básicos
Combining procedural generation with Neural Style Transfer enhances terrain map versatility and realism.
Resumen
In this study, a new technique for generating terrain maps is introduced, utilizing procedural generation and Neural Style Transfer. The method involves creating noise maps using smoothed Gaussian noise or the Perlin algorithm, then applying enhanced Neural Style Transfer with real-world height maps. This fusion of algorithmic generation and neural processing produces diverse terrains aligned with real-world landscapes. The approach offers low computational cost, customized map creation, and accurate replication of terrain morphology. The study highlights the potential of style transfer in transferring morphological information through neural methods.
Estadísticas
Our method completes the process in 2 minutes and 46 seconds.
Training a GAN model for terrain generation can take up to 16 hours with a cluster of GPUs.
The SSIM values show improved similarity between transferred images and original terrains.
Citas
"We consider our approach to be a viable alternative to competing generative models."
"Our findings demonstrate the feasibility of transferring terrain morphological information through Neural Style Transfer."
"The generated images have a higher fidelity to the original real-world maps compared to purely procedurally generated sources."