Siting Zhu · Renjie Qin · Guangming Wang · Jiuming Liu · Hesheng Wang
Table of Contents
sem_guass_slam has been benchmarked with Python 3.10, Pytorch 1.12.1 & CUDA=11.6. The simplest way to install all dependences is to use anaconda and pip in the following steps:
conda create -n sem_gauss python=3.10
conda activate sem_gauss
conda install -c "nvidia/label/cuda-11.6.0" cuda-toolkit
conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.6 -c pytorch -c conda-forge
pip install -r requirements.txtWe will use the replica dataset as an example to show how to use sem_guass_slam. The following steps are similar for other datasets.
To run sem_guass_slam, please use the following command:
python sem_gauss.py configs/replica/replica.pyYou should download the corresponding pth file of the dinov2 in here before running the command above.
To see the evaluation of the reconstructed mesh, please use the following command:
python eval_mesh/mesh_eval.pyYou should rewrite the path of the reconstructed mesh flie and true mesh file in eval_mesh/mesh_eval.py. And you can download the true mesh file in here.
Dataroot is ./data0/replica by default. Please change the input_folder path in the scene-specific config files if datasets are stored somewhere else on your machine.
Download the replica data on this website:replica. Note that the Replica data is generated by the authors of iMAP (but hosted by the authors of NICE-SLAM). Please cite iMAP if you use the data.
Please follow the data downloading procedure on the ScanNet website, and extract color/depth frames from the .sens file using this code.
We thank the authors of the following repositories for their open-source code:
- 3D Gaussians
- Dataloaders
- Baselines
If you find our paper and code useful, please cite us:
@article{zhu2024semgauss,
title={Semgauss-slam: Dense semantic gaussian splatting slam},
author={Zhu, Siting and Qin, Renjie and Wang, Guangming and Liu, Jiuming and Wang, Hesheng},
journal={arXiv preprint arXiv:2403.07494},
year={2024}
}