Latent Radiance Fields with 3D-aware 2D Representations

TL;DR

This paper introduces a novel framework that integrates 3D awareness into 2D latent representations, enabling photorealistic 3D reconstruction from 2D features with improved consistency and generalization.

Contribution

The work presents a three-stage approach combining correspondence-aware autoencoding, latent radiance fields, and VAE-RF alignment to enhance 3D reconstruction from 2D latent spaces.

Findings

Outperforms state-of-the-art in synthesis quality

Demonstrates strong cross-dataset generalization

Achieves photorealistic 3D reconstructions from 2D latent representations

Abstract

Latent 3D reconstruction has shown great promise in empowering 3D semantic understanding and 3D generation by distilling 2D features into the 3D space. However, existing approaches struggle with the domain gap between 2D feature space and 3D representations, resulting in degraded rendering performance. To address this challenge, we propose a novel framework that integrates 3D awareness into the 2D latent space. The framework consists of three stages: (1) a correspondence-aware autoencoding method that enhances the 3D consistency of 2D latent representations, (2) a latent radiance field (LRF) that lifts these 3D-aware 2D representations into 3D space, and (3) a VAE-Radiance Field (VAE-RF) alignment strategy that improves image decoding from the rendered 2D representations. Extensive experiments demonstrate that our method outperforms the state-of-the-art latent 3D reconstruction approaches in terms of synthesis performance and cross-dataset generalizability across diverse indoor and outdoor scenes. To our knowledge, this is the first work showing the radiance field representations constructed from 2D latent representations can yield photorealistic 3D reconstruction performance.