Problem
Research questions and friction points this paper is trying to address.
3D Gaussian Splatting
Inpainting
multi-view inconsistency
3D editing
WINGS: Reference-Free Gaussian Splatting Inpainting with 3D-Native Generative Priors
🤖 AI Summary
This study addresses the limitations of existing 3D Gaussian Splatting inpainting methods, which rely on 2D reference views and consequently suffer from multi-view inconsistency and computationally expensive optimization. To overcome these challenges, this work proposes the first reference-free, 3D-native Gaussian Splatting inpainting framework. The proposed method operates within the embedding space of a pretrained 3D generative prior, integrating a structure completion network to directly reconstruct both the geometry and appearance of missing regions, thereby achieving genuine 3D-native content generation. By fundamentally eliminating multi-view inconsistencies, the framework attains significantly faster inference speeds compared to conventional 2D-based approaches. Extensive experiments and user studies validate the effectiveness and superiority of the proposed method.
📝 Abstract
Inpainting 3D Gaussian Splatting scenes, a key challenge in 3D editing, requires generating plausible content within a masked region of 3D space. Prior approaches rely on 2D diffusion models to produce one or several inpainted reference views, making them susceptible to challenges associated with multi-view inconsistency and lengthy optimization times. Departing from these approaches, we introduce a reference-free Gaussian splatting inpainting method operating natively in 3D. Our method leverages the embedding space of a large, pre-trained 3D prior, combined with a structure completion network to feed a generative prior which reconstructs the missing region's geometry and appearance. Performing content generation entirely in 3D, it avoids the need to reconcile inconsistencies of multiple inpainted reference images, and is faster than related 2D-based methods. We demonstrate the effectiveness of our method qualitatively and quantitatively, through extensive experiments and a user study. To the best of our knowledge, this work is the first Gaussian splatting inpainting method to operate in the learned representation space of a 3D-native generative prior without relying on inpainted reference views.
Innovation
Methods, ideas, or system contributions that make the work stand out.
3D Gaussian Splatting Inpainting
Reference-Free
3D-Native Generative Priors
Structure Completion Network
Multi-view Consistency
🔎 Similar Papers
No similar papers found.
