SuperGaussian: Repurposing Video Models for 3D Super Resolution

TL;DR

This paper introduces SuperGaussian, a method that repurposes existing video models for 3D super-resolution, producing high-quality, 3D-consistent Gaussian Splat models without requiring large 3D training datasets.

Contribution

It presents a novel approach to adapt pretrained video upsampling models for 3D super-resolution, enabling high-quality 3D model generation with minimal additional training.

Findings

Significant improvement in 3D model fidelity across diverse inputs

Effective combination of video upsampling with 3D consolidation for consistency

Category-agnostic method easily integrated into existing workflows

Abstract

We present a simple, modular, and generic method that upsamples coarse 3D models by adding geometric and appearance details. While generative 3D models now exist, they do not yet match the quality of their counterparts in image and video domains. We demonstrate that it is possible to directly repurpose existing (pretrained) video models for 3D super-resolution and thus sidestep the problem of the shortage of large repositories of high-quality 3D training models. We describe how to repurpose video upsampling models, which are not 3D consistent, and combine them with 3D consolidation to produce 3D-consistent results. As output, we produce high quality Gaussian Splat models, which are object centric and effective. Our method is category agnostic and can be easily incorporated into existing 3D workflows. We evaluate our proposed SuperGaussian on a variety of 3D inputs, which are diverse both in terms of complexity and representation (e.g., Gaussian Splats or NeRFs), and demonstrate that our simple method significantly improves the fidelity of the final 3D models. Check our project website for details: supergaussian.github.io