Geometry-Grounded Gaussian Splatting

March 16, 2026 ยท View on GitHub

Baowen Zhang, Chenxing Jiang, Heng Li, Shaojie Shen, Ping Tan

Project Page
Teaser

1. Installation

The code has been tested on Ubuntu 24.04 with CUDA 13.0 (driver 580.126.09), and on Ubuntu 20.04 with CUDA 12.8 (driver 570.133.07).

Clone the repository

git clone https://github.com/HKUST-SAIL/Geometry-Grounded-Gaussian-Splatting.git --recursive
cd Geometry-Grounded-Gaussian-Splatting

Create the environment

conda create -n gggs python=3.12
conda activate gggs

Install PyTorch + dependencies

pip install torch torchvision --index-url https://download.pytorch.org/whl/cu130
pip install -r requirements.txt

Install submodules

pip install submodules/diff-gaussian-rasterization --no-build-isolation
pip install submodules/warp-patch-ncc --no-build-isolation
pip install submodules/simple-knn --no-build-isolation
pip install git+https://github.com/rahul-goel/fused-ssim/ --no-build-isolation

Marching Tetrahedra (tetra-triangulation)

conda install -y conda-forge::cgal
pip install submodules/tetra_triangulation --no-build-isolation

2. Data Preparation

DTU

We train on the preprocessed DTU dataset from 2DGS:
https://surfsplatting.github.io/

For geometry evaluation, download the official DTU point clouds and place them under:

dtu_eval/Offical_DTU_Dataset

DTU dataset page: https://roboimagedata.compute.dtu.dk/?page_id=36

Tanks and Temples (TnT)

Please follow PGSR to preprocess the TnT dataset. For evaluation, download the GT point clouds, camera poses, alignments, and crop files from:
https://www.tanksandtemples.org/download/

Expected structure:

GT_TNT_dataset/
  Barn/
    images/
      000001.jpg
      000002.jpg
      ...
    sparse/
      0/
        ...
    Barn.json
    Barn.ply
    Barn_COLMAP_SfM.log
    Barn_trans.txt
  Caterpillar/
    ...

3. Training & Evaluation

Below are example commands for training, mesh extraction, rendering, and evaluation.

DTU

# Training
python train.py -s <path_to_dtu> -m <output_dir> -r 2 --use_decoupled_appearance 3

# Mesh extraction
python mesh_extract.py -m <output_dir>

# Evaluation
python evaluate_dtu_mesh.py -m <output_dir>

Tanks and Temples (TnT)

# Training
python train.py -s <path_to_preprocessed_tnt> -m <output_dir> -r 2 --use_decoupled_appearance 3


# Mesh extraction
# Tips:
#   - Add --move_cpu to reduce GPU memory usage (if needed).
#   - Add --export_color to export the mesh with vertex colors.
python mesh_extract_tetrahedra.py -m <output_dir> 
# python mesh_extract_tetrahedra.py -m <output_dir> --move_cpu --export_color

# Evaluation
python eval_tnt/run.py \
  --dataset-dir <path_to_gt_tnt> \
  --traj-path <path_to_COLMAP_SfM.log> \
  --ply-path <output_dir>/recon_post.ply \
  --out-dir <output_dir>/mesh

Novel View Synthesis

# Training
python train.py -s <path_to_dataset> -m <output_dir> --eval

# Rendering
python render.py -m <output_dir>

# Evaluation
python metrics.py -m <output_dir>

(Optional) With Spherical Gaussians

python train.py -s <path_to_dataset> -m <output_dir> --eval --sh_degree 2 --sg_degree 7
python render.py -m <output_dir>
python metrics.py -m <output_dir>

4. Viewer

The viewer is based on the original 3D Gaussian Splatting (SIBR) viewer, with minor updates for newer library versions and for loading 3D Gaussian models.

The current version uses embree4 and a newer version of Boost; if you need to build with older library versions, please refer to SIBR_viewers/SIBR_viewers.patch for the required changes.

Build and use it the same way as:
https://github.com/graphdeco-inria/gaussian-splatting

5. Acknowledgements

This project builds on Gaussian Splatting and RaDe-GS:

We refer to gsplat for its use of warp-level reduction to accelerate the backward pass.
https://github.com/nerfstudio-project/gsplat

We integrate: