GSORB-SLAM: Gaussian Splatting SLAM benefits from
ORB features and Transmittance information

💡 News

• [14 July 2025] 🎉 Our paper GSORB-SLAM has been accepted to IEEE RA-L 2025!
• 🌟 If you find this project interesting, please consider starring it to support us!
• 🚧 The code is still under development. Feedback and issue reports are welcome.
• 🙏 Many thanks to all researchers who contribute to the open-source community — respect!

🛠️ Installation

Our project has been tested on Ubuntu 20.04 and 22.04 with CUDA 11.8.

Clone the repository and create the conda environment:

git clone --recurse https://github.com/Aczheng-cai/GSORB-SLAM.git GSORB_SLAM
conda create -n gsorbslam python=3.10
conda activate gsorbslam
conda install -c "nvidia/label/cuda-11.8.0" cuda-toolkit
conda install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=11.8 -c pytorch -c nvidia
pip install -r requirements.txt

# You need to configure the Conda environment in the CMakeLists file.
set(ENVIRONMENT_DIR your_anconda3_path/envs/gsorbslam)

Libtorch (C++)

# In a Terminal
wget https://download.pytorch.org/libtorch/cu118/libtorch-cxx11-abi-shared-with-deps-2.1.2%2Bcu118.zip -O libtorch-cu118.zip
unzip libtorch-cu118.zip -d ~/GSORB_SLAM/Thirdparty/
rm libtorch-cu118.zip

OpenCV with opencv_contrib and CUDA

Our implementation has been tested with OpenCV 4.5.5 and OpenCV 4.9.0. In this document, we take OpenCV 4.9.0 as the default example.

Find the corresponding versions in both the OpenCV realeases and opencv_contrib repositories (e.g., OpenCV 4.9.0 and opencv_contrib 4.9.0, download them into the same directory (e.g., ~/opencv), and extract them. Then, open a terminal and run:

cd ~/opencv
cd opencv-4.9.0/
mkdir build && cd build

# The build options we used in our tests:
cmake -DCMAKE_BUILD_TYPE=RELEASE -DWITH_CUDA=ON -DWITH_CUDNN=ON -DOPENCV_DNN_CUDA=ON -DWITH_NVCUVID=ON -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-11.8 -DOPENCV_EXTRA_MODULES_PATH="../../opencv_contrib-4.9.0/modules" -DBUILD_TIFF=ON -DBUILD_ZLIB=ON -DBUILD_JASPER=ON -DBUILD_CCALIB=ON -DBUILD_JPEG=ON -DWITH_FFMPEG=ON ..

make -j$(nproc)
sudo make install

Eigen3

Required by g2o (see below). Download and install instructions can be found at: http://eigen.tuxfamily.org.

Thirdparty

Install third-party libraries, including Gaussian Splatting, tinyply, yaml, and wandb (more may be added later ⏳).

#install tinyply
cd ~/GSORB_SLAM/Thirdparty/diff_gaussian_rasterization/
conda activate gsorbslam
python setup.py install

#install tinyply
cd ../tinyply
mkdir build && cd build
cmake ..
make -j$(nproc)
sudo make install

#install yaml
cd ../../yaml-cpp
mkdir build && cd build
cmake ..
make -j$(nproc)
sudo make install

#install pangolin-0.6. We recommend using version 0.6.
cd ../../
git clone --recursive https://github.com/stevenlovegrove/Pangolin.git Pangolin
cd Pangolin
git checkout v0.6
mkdir build && cd build
cmake ..
make -j$(nproc)
sudo make install

Installation of GSORB-SLAM🎉️

cd ~/GSORB_SLAM
chmod +x ./build.sh
sh build.sh

📈 Evaluation dependencies（Option）

This step can be skipped if you don't intend to run evaluation directly in C++ after execution. Note: You need to set enable under the Evaluation section in the YAML file to false. 💡Don't worry — we also provide a Python version of the offline evaluation (replay.py).

# Download the LPIPS model and convert the .pth file to .pt.
cd ~/GSORB_SLAM/scripts/
python gen_eval_model.py

# Note
Evalution:
  enable: true   <----If you do not wish to evaluate the results directly, please modify this setting.

🧾Datasets

TUM-RGBD

bash srcipts/dataset_utils/download_tum.sh

Replica

bash scripts/dataset_utils/download_replica.sh

Scannet

Please follow the data downloading procedure on the ScanNet website, and extract color/depth frames from the .sens file using this code.

datasets
├── Scannet
│   ├── scene0000_00
│   │   ├── cameras.json
│   │   ├── color
│   │   ├── depth
│   │   ├── intrinsic
│   │   ├── pose
│   │   └── groundtruth.txt  <----bash pose2traj.sh
├── Replica
└── TUM_RGBD

1. ScanNet requires using the script pose2traj.sh to convert poses into 4×4 continue trajectory matrices for evaluation.

   bash scripts/dataset_utils/pose2traj.sh
   

🚀 Run

Run the TUM RGB-D dataset sequence rgbd_dataset_freiburg1_desk.

cd ~/GSORB_SLAM
./Examples/RGB-D/rgbd_tum ~/GSORB_SLAM/Vocabulary/ORBvoc.txt ~/GSORB_SLAM/Examples/RGB-D/tum/TUM1.yaml ~/datasets/rgbd_dataset_freiburg1_desk/ ~/GSORB_SLAM/Examples/RGB-D/associations/fr1_desk.txt

💡 We provide a script (located at ~/GSORB_SLAM/scripts/run_*.sh) to run the process—simply update the dataset path.

bash scripts/run_tum.sh

Run all datasets bash run.sh. The results are saved in ~/GSORB_SLAM/experiments.

📊Evaluation

You need to modify the YAML file: Dataset.name specifies the result folder, and Dataset.path specifies the ground truth path. Then,

python scripts/replay.py --yamlPath "your_yaml_path"

If you are using the TUM dataset, you also need to add the --tumAss parameter.

🎓Citation

If you find our code/work useful in your research, please consider citing the following:

@ARTICLE{11091447,
  author={Zheng, Wancai and Yu, Xinyi and Rong, Jintao and Ou, Linlin and Wei, Yan and Zhou, Libo},
  journal={IEEE Robotics and Automation Letters}, 
  title={GSORB-SLAM: Gaussian Splatting SLAM benefits from ORB features and Transmittance information}, 
  year={2025},
  pages={1-8},
  doi={10.1109/LRA.2025.3592066}}