This repository contains the official implementation of the paper "Bootstraping Clustering of Gaussians for View-consistent 3D Scene Understanding", an unsupervised semantic-embedded 3DGS framework that achieves view-consistent 3D scene understanding without the need for 2D labels. For more details, please refer to: [Paper].
-[25-05-18] We released the main code of FreeGS.
- TODO
- Setup
- Training
- Evaluation
- Citation
- Acknowledgments
- Release the code for training on the LERF-Mask dataset.
- Release the evaluation code.
- Release the code on other datasets.
- Clone the repository
git clone https://github.com/wb014/FreeGS && cd FreeGS- Setup environment
conda create -n freegs python=3.10
conda activate freegs
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 environment.txt
# install msplat for rasterization
git clone https://github.com/pointrix-project/msplat.git --recursive
cd msplat
pip install .
# install cuml following https://docs.rapids.ai/install/- Download dataset: LERF-Mask.
-
Step 1: Train Gaussians for scene reconstruction based on mini-splatting.
-
Step 2: Train the FreeGS:
python train_freegs.py -s datasets/lerf_mask/${scenename} -m outputs/${scenename} --eval --start_checkpoint outputs/${scenename}/chkpnt30000.pth --exp_name ${expname}TODOIf you find FreeGS helpful, please consider giving this repository a star and citing:
@inproceedings{zhang2025bootstraping,
title={Bootstraping clustering of gaussians for view-consistent 3d scene understanding},
author={Zhang, Wenbo and Zhang, Lu and Hu, Ping and Ma, Liqian and Zhuge, Yunzhi and Lu, Huchuan},
booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
volume={39},
number={10},
pages={10166--10175},
year={2025}
}We thank 3DGS, LangSplat, MSplat, Mini-Splatting, FeatUp for their efforts.