SceneWiz3D

Abstract

We witness significant breakthroughs in the technology for generating 3D objects from text. Existing approaches either leverage large text-to-image models to optimize a 3D representation or train 3D generators on object-centric datasets. Generating entire scenes, however, remains very challenging as a scene contains multiple 3D objects, diverse and scattered. In this work, we introduce SceneWiz3D – a novel approach to synthesize high fidelity 3D scenes from text. We marry the locality of objects with globality of scenes by introducing a hybrid 3D representation – explicit for objects and implicit for scenes. Remarkably, an object, being represented explicitly, can be either generated from text using conventional text-to-3D approaches, or provided by users. To configure the layout of the scene and automatically place objects, we apply Particle Swarm Optimization technique during the distillation process. Furthermore, in the text-to-scene scenario it is difficult for certain parts of the scene (e.g., corners, occlusion) to receive multi-view supervision, leading to inferior geometry. To mitigate the lack of such supervision, we incorporate an RGBD panorama diffusion model, resulting in high quality geometry. Extensive evaluation supports that our approach achieves superior quality over previous approaches, enabling the generation of detailed and view-consistent 3D scenes

Overview

Our goal is to create high-fidelity 3D scenes from text. We propose a hybrid scene representation where objects of Interest (OOIs) are modeled by Deep Marching Tetrahedra. The remaining parts of the scene is modeled by Neural Radiance Field (NeRF).

As we disentangle OOIs from the rest of the scene in our hybrid representation, we need to determine the configuration for each object, including coordinates, scaling factor, and rotation degree. We find that it is nontrivial to adopt gradient descent to update the low-dimensional config, but instead propose to use Particle Swarm Optimization (PSO) to automatically configure the scene's layout. Different from prior works that only employ perspectivew view distillation, we also incorporate LDM3D, a diffusion model finetuned on panoramic images in RGBD space to provide additional guidance. During the optimization process, LDM3D provides additional prior information: The RGBD knowledge yield supervision in depth, while panoramic knowledge mitigates the issues of limited views with perspective images and disambiguates the global structure of a scene.