GaME: Gaussian Mapping for Evolving Scenes

CVPR 2026

Vladimir Yugay* · Kersten Thies* · Luca Carlone · Theo Gevers · Martin Oswald · Lukas Schmid

Project Page

⚙️ Setting Things Up

Clone the repo:

git clone https://github.com/VladimirYugay/GaME.git

We tested the installation with gcc and g++ of versions 10, 11 and 12. Also, make sure that nvcc --version matches nvidia-smi version.

Run the following commands to set up the environment

conda create -n game python=3.10 -y
conda activate game

Install pytorch:

# CUDA 11.8
conda install pytorch==2.5.1 torchvision==0.20.1 pytorch-cuda=11.8 -c pytorch -c nvidia
# CUDA 12.1
conda install pytorch==2.5.1 torchvision==0.20.1 pytorch-cuda=12.1 -c pytorch -c nvidia
# CUDA 12.4
conda install pytorch==2.5.1 torchvision==0.20.1 pytorch-cuda=12.4 -c pytorch -c nvidia

Install other dependencies:

conda install -c conda-forge faiss-gpu=1.8.0 git-lfs
pip install ./submodules/diff-gaussian-rasterization/ --no-build-isolation
pip install ./submodules/flashsplat-rasterization/ --no-build-isolation
pip install ./submodules/simple-knn/ --no-build-isolation
pip install -r requirements.txt

🔨 Running GaME

Here we elaborate on how to load the necessary data, configure GaME for your use-case, debug it, and how to reproduce the results mentioned in the paper.

Getting the Data

We tested our code on Flat and Aria datasets. Make sure to install git lfs and hugging face cli before proceeding.

Flat was created by Panoptic Mapping authors. However, it is a bit tricky to find it on the web. Therefore, we uploaded it to HF datasets for easier access. Install git lfs and download it by running:
git lfs install
git clone https://huggingface.co/datasets/voviktyl/GaME_Flat

Aria consists of clips from AriaDigitalTwin. The clips are covering the area that undergoes changes outside of the camera view. You can download already processed data with the command:
git clone https://huggingface.co/datasets/voviktyl/GaME_Aria

Running the code

python run.py --config_path configs/<dataset_name>/<config_name> --data_path <path_to_the_scene> --output_path <output_path>

For example:

python run.py --config_path configs/AriaMultiagent/room0.yaml --data_path <path_to>/AriaMultiagent/room0 --output_path output/AriaMultiagent/room0

Check the configs to set up wandb for loggin.

Reproducing Results

While we tried to make all parts of our code deterministic, differential rasterizer of Gaussian Splatting is not. The metrics can be slightly different from run to run.

If you are running on a SLURM cluster, you can reproduce the results for all the datasets by running the files in the scripts folder. For example, for Aria dataset:

./scripts/reproduce_aria_sbatch.sh

🙏 Acknowledgments

Mapping module is based on Gaussian-SLAM. We thank the authors of FlashSplat for their work.

📌 Citation

If you find our paper and code useful, please cite us:

@misc{yugay2025gaussianmappingevolvingscenes,
      title={Gaussian Mapping for Evolving Scenes}, 
      author={Vladimir Yugay and Thies Kersten and Luca Carlone and Theo Gevers and Martin R. Oswald and Lukas Schmid},
      year={2025},
      eprint={2506.06909},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2506.06909}, 
}