Accepted by ICRA 2025. 🔗 Project Website
git clone https://github.com/RPL-CS-UCL/gs_localization.git --recursive
- create an environment
conda create -n gsloc python=3.9
conda activate gsloc
- install pytorch and other dependencies.
for cuda11.6
pip install torch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 --index-url https://download.pytorch.org/whl/cu116
pip install -r requirements.txt
for cuda11.8
pip install torch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt
- install submodules
pip install submodules/diff-gaussian-rasterization
pip install submodules/simple-knn/
The following structure shows how the dataset folder is organized. Please ensure your local datasets/ directory follows this layout:
gs-localization/
└── datasets/
├── 7scenes/
│ ├── chess/
│ └── seq-01/
│ ├── frame-000000.color.png
│ ├── ...
│ └── frame-000XXX.color.png
│ ├── fire/
│ ├── heads/
│ ├── office/
│ ├── pumpkin/
│ ├── redkitchen/
│ ├── stairs/
│ ├── 7scenes_densevald_retrieval_top_10/
│ ├── 7scenes_sfm_triangulated/
│ └── train_fewshot_all
├── 7scenes_additional/
│ └── chess/
│ └── seq-01/
│ ├── frame-000000.depth_cali.png
│ ├── ...
│ └── frame-000XXX.depth_cali.png
├── cambridge/
├── cambridge_additional/
├── 360_v2/
└── nerf_llff_data/
Here is how LoGS re-localizes the scenes we experimented in our paper, e.g. 7-scenes (dslam ground truth and full training images). You need to run four python files one by one.
- pre-process the 7-scenes dataset.
python gs_localization/process/train_test_split_7scenes_full_dslam.py
- obtain a SfM point could of training images and rough initial poses of testing images through PnP-RANSAC.
python gs_localization/sfm/7scenes_sfm_full_dslam.py
- train a 3DGS map of training images.
python gs_localization/gs/7scenes_gs_full_dslam.py
- refines poses with gradient decent and we are done.
python gs_localization/pipelines/7scenes_localize_full_dslam.py
@inproceedings{cheng2025logs,
title = {LoGS: Visual Localization via Gaussian Splatting with Fewer Training Images},
author = {Cheng, Yuzhou and Jiao, Jianhao and Wang, Yue and Kanoulas, Dimitrios},
booktitle = {International Conference on Robotics and Automation (ICRA)},
pages = {},
year = {2025},
organization = {IEEE},
dimensions = {true},
}This work was supported by the UKRI FLF [MR/V025333/1] (RoboHike).
For the purpose of Open Access, the author has applied a CC BY public copyright license to any Author Accepted Manuscript version arising from this submission.
Prof.Dimitrios Kanoulas is also with Archimedes/Athena RC, Greece.