Gaussian splat completion improves 3D object restoration from images
GSComplete: Gaussian Splat Completion with 2D Diffusion Priors
Computer Vision and Pattern Recognition
Summary
3D objects made from tiny Gaussian shapes often have gaps when their data is incomplete. The authors introduce a way to fill in these missing parts while keeping the original visible pieces intact. They use existing 2D image knowledge to guide this process instead of needing 3D training data. Their method is tested on a new dataset and preserves the original object's details more accurately than previous ways.
What this means in practice
- •For 3d graphics designers: Improve 3D object restoration workflows by automatically completing partial Gaussian splat models without losing visible original details.
- •For augmented reality developers: Generate complete high-fidelity 3D models from incomplete scans to enhance visual quality in AR experiences.$Commercial implications: Enables selling AR content generation tools that restore partial 3D scans automatically for better immersion.
Authors
Elias Brugger, Philipp Erler, Stefan Ohrhallinger, Paul Guerrero
Abstract
Gaussian splats provide a fast, high-fidelity representation for 3D objects but are often constructed from incomplete input data in practice, leaving missing regions. Existing completion methods either do not preserve the original splats or require scarcely available 3D training data. We propose GSComplete, which combines 3D generation based on Score Distillation Sampling with a novel preservation loss that encourages the original splats to be preserved where they should be visible. This effectively completes the Gaussian splat object using only 2D diffusion priors while fully preserving existing splats and generating new splats only in missing regions, without occluding the input. To evaluate our approach, we introduce a new dataset of partial Gaussian splat objects and show that GSComplete achieves significantly more accurate preservation of the input than existing methods with comparable plausibility of the completed result. Our code and dataset will be made available upon acceptance.