This paper focuses on achieving the fusion of images and point clouds to enable coarse visual localization of a single image within a pre-built point cloud map.
git clone https://github.com/whu-lyh/SaliencyI2PLoc.git --recursive
cd scripts
bash install.shYou may required to change the coding manner of sh files using
sed -i "s/\r//" *.shto avoid the file unrecognition.
- both pytorch1.13.1-cuda11.6 and pytorch2.1.2-cuda12.1 works
The model weights and tha datasets could be downloaded from GoogleDrive. The pretrained models of ResNet and ViT used in our job could be download at here.
cd scripts
bash train.shThe configuration information will be loaded all in once from the CrossModalityRetrieval.yaml style file, including the optimizer, scheduler, dataset, model and other configuration.
cd scripts
bash test.shAdjust the test data sequences that you want to test at /config/dataset_configs folder.
The details of the used model can be found in Architectures.md.
The details of the used datasets can be found in Datasets.md.
If you find our work is useful to yours, please cite our paper.
@article{LI2025103015,
title = {SaliencyI2PLoc: Saliency-guided image–point cloud localization using contrastive learning},
journal = {Information Fusion},
volume = {118},
pages = {103015},
year = {2025},
issn = {1566-2535},
doi = {https://doi.org/10.1016/j.inffus.2025.103015}
}