MrNeRF (@janusch_patas)
2024-10-22 | โค๏ธ 144 | ๐ 20
[NIPS โ24 Spotlight] 3DGS-Enhancer: Enhancing Unbounded 3D Gaussian Splatting with View-consistent 2D Diffusion Priors
Contributions (cited):
-
To the best of our knowledge, this is the first work to tackle the problem of enhancing low-quality 3DGS rendering results, an issue that widely exists in practical 3DGS applications.
-
We propose a novel pipeline 3DGS-Enhancer that addresses the 3DGS enhancement problem. 3DGS-Enhancer reformulates the 3D-consistent image restoration task as temporally consistent video generation, such that powerful video LDMs can be leveraged for generating both high-quality and 3D-consistent images.
Novel 3DGS fine-tuning strategies are also devised for an effective integration of the enhanced views with the original 3DGS representation.
- We conduct extensive experiments on large-scale datasets of unbounded scenes to demonstrate the effectiveness of the proposed methods over existing state-of-the-art few-shot NVS methods.
๋ฏธ๋์ด
