/DenseUAV

「TIP」Vision-Based UAV Self-Positioning in Low-Altitude Urban Environments

Primary LanguagePythonApache License 2.0Apache-2.0

Vision-Based UAV Self-Positioning in Low-Altitude Urban Environments

This repository contains code and dataset for the paper titled Vision-Based UAV Self-Positioning in Low-Altitude Urban Environments. In this paper, we propose a method for accurately self-positioning unmanned aerial vehicles (UAVs) in challenging low-altitude urban environments using vision-based techniques. We provide the DenseUAV dataset and a Baseline model implementation to facilitate research in this task. Thank you for your kind attention.

News

  • 2023/12/18: Our paper is accepted by IEEE Trans on Image Process.
  • 2023/8/14: Our dataset and code are released.

Table of contents

About Dataset

The dataset split is as follows:

Subset UAV-view Satellite-view Classes universities
Training 6,768 13,536 2,256 10
Query 2,331 4,662 777 4
Gallery 9099 18198 3033 14

More detailed file structure:

├── DenseUAV/
│   ├── Dense_GPS_ALL.txt           /* format as: path latitude longitude height
│   ├── Dense_GPS_test.txt
│   ├── Dense_GPS_train.txt
│   ├── train/
│       ├── drone/                   /* drone-view training images
│           ├── 000001
│               ├── H100.JPG
│               ├── H90.JPG
│               ├── H80.JPG
|           ...
│       ├── satellite/               /* satellite-view training images
│           ├── 000001
│               ├── H100_old.tif
│               ├── H90_old.tif
│               ├── H80_old.tif
│               ├── H100.tif
│               ├── H90.tif
│               ├── H80.tif
|           ...
│   ├── test/
│       ├── query_drone/             /* UAV-view testing images
│       ├── query_satellite/         /* satellite-view testing images

Prerequisites

  • Python 3.7+
  • GPU Memory >= 8G
  • Numpy 1.21.2
  • Pytorch 1.10.0+cu113
  • Torchvision 0.11.1+cu113

Installation

It is best to use cuda version 11.3 and pytorch version 1.10.0. You can download the corresponding version from this website and install it through pip install. Then you can execute the following command to install all dependencies.

pip install -r requirments.txt

Create the directory for saving the training log and ckpts.

mkdir checkpoints

Dataset & Preparation

Download DenseUAV upon request. You may use the request Template.

Train & Evaluation

Training and Testing

You could execute the following command to implement the entire process of training and testing.

bash train_test_local.sh

The setting of parameters in train_test_local.sh can refer to Get Started.

Evaluation

The following commands are required to evaluate Recall and SDM separately.

cd checkpoints/<name>
python test.py --name <name> --test_dir <dir/to/testdir/of/dataset> --gpu_ids 0 --num_worker 4

the <name> is the dir name in your training setting, you can find in the checkpoints/.

For Recall

python evaluate_gpu.py

For SDM

python evaluateDistance.py --root_dir <dir/to/root/of/dataset>

We also provide the baseline checkpoints, quark one-drive.

unzip <file.zip> -d checkpoints
cd checkpoints/baseline
python test.py --test_dir <dataset_root>/test
python evaluate_gpu.py
python evaluateDistance.py --root_dir <dataset_root>

Supported Methods

Augment Backbone Head Loss
Random Rotate ResNet MaxPool CrossEntropy Loss.
Random Affine EfficientNet AvgPool Focal Loss
Random Brightness ConvNext MaxAvgPool Triplet Loss
Random Erasing DeiT GlobalPool Hard-Mining Triplet Loss
PvT GemPool Same-Domain Triplet Loss
SwinTransformer LPN Soft-Weighted Triplet Loss
ViT FSRA KL Loss

License

This project is licensed under the Apache 2.0 license.

Citation

The following paper uses and reports the result of the baseline model. You may cite it in your paper.

@ARTICLE{DenseUAV,
  author={Dai, Ming and Zheng, Enhui and Feng, Zhenhua and Qi, Lei and Zhuang, Jiedong and Yang, Wankou},
  journal={IEEE Transactions on Image Processing},
  title={Vision-Based UAV Self-Positioning in Low-Altitude Urban Environments},
  year={2024},
  volume={33},
  number={},
  pages={493-508},
  doi={10.1109/TIP.2023.3346279}}

Related Work