High-fidelity 3D GAN Inversion by Pseudo-multi-view Optimization

TL;DR

This paper introduces a novel 3D GAN inversion method that achieves high-fidelity, photo-realistic view synthesis and detailed image editing from a single image by leveraging pseudo-multi-view estimation and visibility analysis.

Contribution

It proposes a new pipeline for 3D GAN inversion that effectively balances geometry and texture preservation using pseudo-multi-view and generative priors, outperforming existing methods.

Findings

Achieves superior reconstruction quality over state-of-the-art methods.

Enables high-fidelity novel view synthesis from a single image.

Supports image attribute editing and 3D-aware texture modification.

Abstract

We present a high-fidelity 3D generative adversarial network (GAN) inversion framework that can synthesize photo-realistic novel views while preserving specific details of the input image. High-fidelity 3D GAN inversion is inherently challenging due to the geometry-texture trade-off in 3D inversion, where overfitting to a single view input image often damages the estimated geometry during the latent optimization. To solve this challenge, we propose a novel pipeline that builds on the pseudo-multi-view estimation with visibility analysis. We keep the original textures for the visible parts and utilize generative priors for the occluded parts. Extensive experiments show that our approach achieves advantageous reconstruction and novel view synthesis quality over state-of-the-art methods, even for images with out-of-distribution textures. The proposed pipeline also enables image attribute editing with the inverted latent code and 3D-aware texture modification. Our approach enables high-fidelity 3D rendering from a single image, which is promising for various applications of AI-generated 3D content.