3D-SceneDreamer: Text-Driven 3D-Consistent Scene Generation

TL;DR

This paper introduces a novel method for text-driven 3D scene generation that enhances global 3D consistency and visual quality by refining local views using a tri-plane NeRF representation and a generative refinement network.

Contribution

The paper proposes a new approach that combines a tri-plane NeRF with a generative refinement network to improve 3D consistency and quality in text-driven scene generation.

Findings

Supports diverse scene generation

Achieves higher visual quality

Ensures better 3D consistency

Abstract

Text-driven 3D scene generation techniques have made rapid progress in recent years. Their success is mainly attributed to using existing generative models to iteratively perform image warping and inpainting to generate 3D scenes. However, these methods heavily rely on the outputs of existing models, leading to error accumulation in geometry and appearance that prevent the models from being used in various scenarios (e.g., outdoor and unreal scenarios). To address this limitation, we generatively refine the newly generated local views by querying and aggregating global 3D information, and then progressively generate the 3D scene. Specifically, we employ a tri-plane features-based NeRF as a unified representation of the 3D scene to constrain global 3D consistency, and propose a generative refinement network to synthesize new contents with higher quality by exploiting the natural image prior from 2D diffusion model as well as the global 3D information of the current scene. Our extensive experiments demonstrate that, in comparison to previous methods, our approach supports wide variety of scene generation and arbitrary camera trajectories with improved visual quality and 3D consistency.