PRGS: Patch-to-Region Graph Search for Visual Place Recognition

Setup

Install the python environment using

pip3 install -r requirements.txt
pip3 install torch==1.12.1+cu113 torchvision==0.13.1+cu113 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu113

Getting Started

This repo follows the Visual Geo-localization Benchmark. You can refer to it (VPR-datasets-downloader) to prepare datasets.

The dataset should be organized in a directory tree as such:

── datasets
    └── pitts30k
        └── images
            ├── train
            │   ├── database
            │   └── queries
            ├── val
            │   ├── database
            │   └── queries
            └── test
                ├── database
                └── queries

Train in MSLS

For using MSLS dataset, you should get directory mapillary_sls from msls to use MSLS datasets. And then you should get DeiT backbone from msls_v2_deits.pth, save to resume directory. Lastly, you should perform pre_compute_mining to get hard negative list.

python pre_compute_mining.py

bash train_reranking.sh

Test in MSLS

You should get our model that trained in msls from msls_train_prgs.pt,save to resume directory.

bash test.sh

Acknowledgements

Parts of this repo are inspired by the following great repositories:

• Visual Geo-localization Benchmark
• VPR-datasets-downloader
• Mapillary