DeepGPS

This repository is the official PyTorch implementation of DeepGPS, which has been published in IEEE Transactions on Mobile Computing (TMC), a premier journal in the field of mobile computing.

If you find DeepGPS useful, please cite our paper with the following bibtex:

@article{liu2024deepgps,
title={{DeepGPS}: deep learning enhanced {GPS} positioning in urban canyons},
author={Liu, Zhidan and Liu, Jiancong and Xu, Xiaowen and Wu, Kaishun},
journal={IEEE Transactions on Mobile Computing},
volume={23},
number={1},
pages={376--392},
year={2024},
publisher={IEEE}
}

Installation

Requirements: Python >= 3.5, Anaconda3

• Install Pytorch >= 1.8.1

pip install torch==1.8.1+cu111 torchvision==0.9.1+cu111 -f https://download.pytorch.org/whl/torch_stable.html

The latest tested combination is: Python 3.8.5 + Pytorch 1.8.1 + cuda 11.1

Quick Start

python Predict.py

Sample Description

The samples in stored as .npz file. which can be accessed by:

import numpy as np
npz_file = np.load('deepgps.npz')

Each .npz file contains contextual information and ground truth for one position point. The content in npz_file is following:

- npz_file['arr_0']：Environment Matrix.
- npz_file['arr_1']：Four float number. (corrected latitude,corrected longitude, original latitude, original longitude)
- npz_file['arr_2']：The number of seconds that have passed since 00:00:00, 20 June 2019.
- npz_file['arr_3']：The relative coordinate of corrected point (The original point is at (50,50)).
- npz_file['arr_4']：Gaussian Peak Representaiton for ground truth.
- npz_file['arr_5']：Skyplot Matrix.

Download

The weights of a well-trained model can be downloaded from Baidu Cloud Storage (百度网盘) using the following information:

链接：https://pan.baidu.com/s/1ehZHT895ass9PEZ3AsFaTQ?pwd=jb0w
提取码：jb0w

Other Works

Please follow our group at https://github.com/SZU-BDUC/.