This repo contains the Implementation and dataset of the paper "Cross-View Image Sequence Geo-localization".

Prerequisite

---

Python 3
PyTorch > 1.10
TorchVision > 0.11
tqdm
numpy

Training

---

python train.py  --save_name SAVE_NAME --batch_size 24 --MHA_layers 6 --nHeads 8 --max_masked 6

Test

---

python test.py --model_folder SAVE_NAME

Dataset

---

Please fill out this form to obtain access to the collected dataset. To be noticed, we will only share the data if it will be used for RESEARCH PURPOSE only. After downloading the dataset, please unzip and put all three folders (json, satellite, and street) under the dataset folder.

Citation

---

@INPROCEEDINGS{10031017,
  author={Zhang, Xiaohan and Sultani, Waqas and Wshah, Safwan},
  booktitle={2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)}, 
  title={Cross-View Image Sequence Geo-localization}, 
  year={2023},
  volume={},
  number={},
  pages={2913-2922},
  keywords={Computer vision;Databases;Image sequences;Global Positioning System;Videos;Applications: Remote Sensing;Image recognition and understanding (object detection;categorization;segmentation;scene modeling;visual reasoning)},
  doi={10.1109/WACV56688.2023.00293}}