AI-Driven Stylization of 3D Environments

TL;DR

This paper presents a novel AI-driven pipeline that stylizes 3D scenes by integrating advanced 3D representations with image stylization and generative models, enabling high-fidelity scene creation.

Contribution

It introduces a new method combining NeRFs, 3D Gaussian Splatting, and image stylization techniques for scene stylization, which was not previously explored.

Findings

Successful stylization of 3D scenes with generated objects

Demonstrated integration of 3D representations with image stylization

Discussion of current limitations and future directions

Abstract

In this system, we discuss methods to stylize a scene of 3D primitive objects into a higher fidelity 3D scene using novel 3D representations like NeRFs and 3D Gaussian Splatting. Our approach leverages existing image stylization systems and image-to-3D generative models to create a pipeline that iteratively stylizes and composites 3D objects into scenes. We show our results on adding generated objects into a scene and discuss limitations.