Single-Stage Diffusion NeRF: A Unified Approach to 3D Generation and Reconstruction

TL;DR

This paper introduces SSDNeRF, a unified single-stage diffusion-based model that jointly learns 3D reconstruction and generative priors for neural radiance fields, enabling high-quality 3D synthesis and reconstruction from limited views.

Contribution

It proposes a novel end-to-end training paradigm for NeRF and diffusion models, eliminating the need for pretraining and enabling joint 3D reconstruction and prior learning.

Findings

Achieves results comparable or superior to task-specific methods.

Effective in sparse-view 3D reconstruction.

Supports unconditional 3D object generation.

Abstract

3D-aware image synthesis encompasses a variety of tasks, such as scene generation and novel view synthesis from images. Despite numerous task-specific methods, developing a comprehensive model remains challenging. In this paper, we present SSDNeRF, a unified approach that employs an expressive diffusion model to learn a generalizable prior of neural radiance fields (NeRF) from multi-view images of diverse objects. Previous studies have used two-stage approaches that rely on pretrained NeRFs as real data to train diffusion models. In contrast, we propose a new single-stage training paradigm with an end-to-end objective that jointly optimizes a NeRF auto-decoder and a latent diffusion model, enabling simultaneous 3D reconstruction and prior learning, even from sparsely available views. At test time, we can directly sample the diffusion prior for unconditional generation, or combine it with arbitrary observations of unseen objects for NeRF reconstruction. SSDNeRF demonstrates robust results comparable to or better than leading task-specific methods in unconditional generation and single/sparse-view 3D reconstruction.