blancocd/Geometrically_Accurate_FeedForward_Gaussian_Splatting_for_Unbounded_Scenes
<p align="center"> <h1 align="center">Master's Thesis: Geometrically Accurate Feed‐Forward Gaussian Splatting for Unbounded Scenes</h1> <p align="center"> <a href="https://blancocd.com/">César Díaz Blanco</a> </p> <p align="center"> Supervisors: <a href="https://niujinshuchong.github.io/">Zehao Yu</a> · <a href="https://haofeixu.github.io/">Haofei Xu</a> </p> <p align="center"> Examiners: <a href="https://www.cvlibs.net/">Andreas Geiger</a> · <a href="https://virtualhumans.mpi-inf.mpg.de/people/pons-moll.html">Gerard Pons-Moll</a> </p> <h3 align="center"> <a href="https://blancocd.com/ThesisWeb/MasterThesisCesarDiazBlanco_large.pdf">Thesis</a> | <a href="https://blancocd.com/ThesisWeb/">Project Page</a> | <a href="https://github.com/blancocd/Geometrically-Accurate-Feed-Forward-Gaussian-Splatting-for-Unbounded-Scenes">Code</a> </h3> </p>
This thesis enables feed-forward Gaussian splatting models to achieve geometrically accurate 3D representations of unbounded scenes. Our method performs on par with costly per-scene optimization and surpasses state-of-the-art depth estimation models.