This project hosts the code for implementing the TIB-Net for drone detection.

TIB-Net: Drone Detection Network With Tiny Iterative Backbone
H. Sun, J. Yang, J. Shen, D. Liang, L. Ning-Zhong and H. Zhouin
In: IEEE Access, vol. 8, pp. 130697-130707, 2020

This implementation is based on EXTD, and the Environment Requirement is basically the same. The following lists our current experimental environment:

• Ubuntu 20.04.1 LTS
• TITAN V
• python 3.6
• pytorch 1.0.0
• CUDA 9.0
• GCC 6

a. Create a conda virtual environment and activate it.

conda create -n tibnet python=3.6  
conda activate tibnet

b. Clone the TIB-Net repository.

git clone https://github.com/kyn0v/TIB-Net.git
cd TIB-Net

c. Install build requirements

pip install -r requirements.txt

It is recommended to symlink the dataset root to $TIB-Net/data. If your folder structure is different, you may need to change the corresponding paths in ./config.py.

├── backbone
├── config.py
├── data
├── dataset 
│   ├── Annotations
│   │   ├── 000001.xml
│   │   ├── ...
│   │   └── 002850.xml
│   ├── JPEGImages
│   │   ├── 000001.jpg
│   │   ├── ...
│   │   └── 002850.jpg
│   ├── test.txt
│   ├── train.txt
│   └── val.txt
├── result
│   ├── demo
│   │   ├── img
│   │   └── img_det
│   ├── detection
│   └── evaluation
├── demo.py
└── ...

In addition, our collected drone dataset has been uploaded to Google Drive, and you can download HERE.

You can use the following commands to test a dataset.

python test.py --weight ${WEIGHT_FILE}

Then the result file (in '.pkl' & 'txt' format) would be save in ./result/detection/.

We provide a demo script to test all images in specified path(default:./result/demo/img/), and save the annotated images(default:./result/demo/img_det/):

python demo.py [--image_dir ${IMGDIR}] [--save_dir ${SAVEDIR}] --weight ${WEIGHT_FILE} [--thresh ${THRESH}]

By default we evaluate the model on the validation set after each epoch, you can change the evaluation interval by modifying the training part in ./train.py.

python train.py [--batch_size ${BATCHSIZE}] [--resume ${CHECKPOINT}] [--num_workers ${WORKSNUM}] [--lr ${LEARNINGRATE}]

NOTE:

• We only support single-GPU training and testing so far.
• --resume loads both the model weights and optimizer status, and the epoch is also inherited from the specified checkpoint.

We provide a useful evaltool under ./evaltool/ directory. You can compute mAP and save the evaluation result would be saved in ./result/evaluation/.

cd evaltool/
python reval_voc.py

Then the result file (in '.pkl' format) would be save in ./result/evaluation/. Moreover, the function about drawing simple PR curve was commented, and you can activate the feature by modifying corresponding file.

We appreciate all the contributors who open source code and promote community development, and we wish that our work could also inspire other researchers.

• TIB-Net: Drone Detection Network With Tiny Iterative Backbone
• EXTD_Pytorch
• S3FD.pytorch

@ARTICLE{9141228,
  author={H. {Sun} and J. {Yang} and J. {Shen} and D. {Liang} and L. {Ning-Zhong} and H. {Zhou}},
  journal={IEEE Access}, 
  title={TIB-Net: Drone Detection Network With Tiny Iterative Backbone}, 
  year={2020},
  volume={8},
  number={},
  pages={130697-130707},}

May the force be with you~