This repository contains the official open-source implementation of the paper "Horseshoe Splatting: Handling Structural Sparsity for Uncertainty-Aware Gaussian-Splatting Radiance Field Rendering". We introduce Horseshoe Splatting, a Bayesian extension of 3D Gaussian Splatting (3DGS) that jointly addresses structured sparsity in per-splat covariances and delivers calibrated uncertainty.

Hardware Requirements

CUDA-ready GPU with Compute Capability 7.0+

Software Requirements

Conda (recommended for easy setup)

C++ Compiler for PyTorch extensions

CUDA SDK 11 for PyTorch extensions

C++ Compiler and CUDA SDK must be compatible

Please clone with submodules (The repository will be public after the paper is accepted)

# SSH
git clone git@github.com:KaedeGo/hs-gs.git --recursive

or

# HTTPS
git clone https://github.com/KaedeGo/hs-gs.git --recursive

We provide conda environment file to creat experiment environment:

conda env create --file environment.yml
conda activate hs_splatting

We test our code on ubuntu system, please refer to original 3DGS repo about the potential error building the environment or running on windows.

The LF dataset and LLFF dataset files are provided here: LF dataset, LLFF dataset.

Please unzip and put them under the a dataset folder:

├──dataset
│   │  
│   ├──── LF
│   └──── nerf_llff_data

To train and evaluate the image quality and the image/depth uncertainty on LF dataset:

sh scripts/train_render_lf.sh

To train and evaluate the image quality and image uncertainty quality on LLFF dataset:

sh scripts/train_render_llff.sh

To perform the active training on LLFF dataset:

sh scripts/active_llff.sh