CHPDet

PyTorch implementation of "Arbitrary-Oriented Ship Detection through Center-Head Point Extraction", [<a href="https://raw.github.com/zf020114/DARDet/master/Figs/GRSL.pdf">pdf</a>].<br><br>

Highlights:

1. we propose a center-head point extraction based detector (named CHPDet) to achieve arbitrary-oriented ship detection in remote sensing images.

<p align="center"> <img src="https://raw.github.com/zf020114/CHPDet/master/Figs/framework.png" width="100%"></p>

2. Our CHPDet achieves state-of-the-art performance on the FGSD2021, HRSC2016 and UCAS-AOD datasets with high efficiency.

<p align="center"> <img src="https://raw.github.com/zf020114/CHPDet/master/Figs/result3.png" width="100%"></p> <p align="center"> <img src="https://raw.github.com/zf020114/CHPDet/master/Figs/resulthrsc.png" width="100%"></p> <p align="center"> <img src="https://raw.github.com/zf020114/CHPDet/master/Figs/resultaod.png" width="100%"></p>

2. we proposed a new dataset named FGSD2021 for multi-class arbitrary-oriented ship detection in remote sensing images at a fixed GSD.

<p align="center"> <img src="https://raw.github.com/zf020114/CHPDet/master/Figs/dataset.png" width="75%"></p>

Benchmark and model zoo (exact code 'nudt')

|Model | Backbone | Dataset | Rotate | img_size | Inf time (fps) | box AP (ori./now) | Download| |:-------------:| :-------------: | :-----:| :-----: | :-----: | :------------: | :----: | :---------------------------------------------------------------------------------------: | |CHPDet | DLA-34_OIM | FGSD2021 | ✓ | 512x512 | 41.7 | 87.91 | model(exact code 'nudt') | |CHPDet | Hourglass_104 | HRSC2016 | ✓ | 1024x1024 | 13.7 | 90.55 | model(exact code 'nudt') |

The FGSD2021 dataset is available at [<a href="https://pan.baidu.com/s/1q1YQsFAR6nvWIVoa6eo85w?pwd=nudt提取码：nudt">FGSD2021</a>] (exact code 'nudt')

The FGAD2022 dataset is available at [<a href="https://pan.baidu.com/s/1PO0ZtHI5-4J9j5tcKSPp6Q?pwd=nudt提取码：nudt">FGAD2022</a>] (exact code 'nudt')

Installation

The code was tested on Ubuntu 16.04, with Anaconda Python 3.7 and PyTorch v1.4.0. NVIDIA GPUs are needed for both training and testing. After installing Anaconda:

1. [Optional but recommended] create a new conda environment.

   conda create --name CHPDet python=3.7
   

   And activate the environment.

   conda activate CHPDet
   

2. Install PyTorch 1.4.0:

3. Install COCOAPI:

   # COCOAPI=/path/to/clone/cocoapi
   git clone https://github.com/cocodataset/cocoapi.git $COCOAPI
   cd $COCOAPI/PythonAPI
   make
   python setup.py install --user
   

4. Clone this repo:

   CHPDet_ROOT=/path/to/clone/CHPDet
   git clone https://github.com/zf020114/CHPDet $CHPDet_ROOT
   

5. Install the requirements

   pip install -r requirements.txt
   

6. Compile deformable convolutions (from DCNv2).

   cd $CHPDet_ROOT/src/lib/models/networks/DCNv2
   ./make.sh
   

7. compile orn from s2anet cd $CHPDet_ROOT/src/lib/models/networks/orn bash make.sh

7.install DOTA_devkit to get the evel result 7.1. install swig

    sudo apt-get install swig

7.2. create the c++ extension for python

    swig -c++ -python polyiou.i
    python setup.py build_ext --inplace

8. [Optional, only required if you are using extremenet or multi-scale testing] Compile NMS if your want to use multi-scale testing or test ExtremeNet.

   cd $CenterNet_ROOT/src/lib/external
   make
   

9. Download pertained models for detection or pose estimation and move them to $CHPDet_ROOT/models/.

Prepare dataset.

Transform dataset to COCO format

1. prepare data as FGSD2021 using the tool labimg2 (https://github.com/chinakook/labelImg2),
Here, we give an example for image and annotation in ./Figs/

2. crop data to fixed size using ./src/0data_crop.py.

3. make json file using ./src/1make_json_anno.py.

4. check the json file by ./src/3show_json_anno.py

5. It is recommended to symlink the dataset root to `data/coco`.

• Inference CHPDet on DOTA.
  1. download model to /exp/multi_pose/dla_USnavy512_RGuass1_12raduis_arf
  2. python test.py multi_pose --exp_id dla_USnavy512_RGuass1_12raduis_arf --dataset coco_hp --resume --debug 2

*If you want to evaluate the result on DOTA test-dev, zip the files in work_dirs/dardet_r50_fpn_1x_dcn_val/result_after_nms and submit it to the evaluation server.

Train a model

1.1. Train centernet-Rbb: python main.py ctdet --exp_id CenR_usnavy512_dla_2x --arch dla_34 --batch_size 26 --master_batch 14--lr 5e-4 --gpus 0,1 --num_workers 8 --num_epochs 230 --lr_step 180,210 --load_model ../exp/multi_pose_dla_3x.pth 1.2. test centernet-Rbb : python test.py ctdet --exp_id CenR_usnavy512_dla_2x --arch dla_34 --keep_res --resume

2.1 Train CHP_DLA_34:

python main.py multi_pose --exp_id dla_USnavy512_RGuass1_12raduis_arf --arch dlaarf_34 --dataset coco_hp --batch_size 28 --master_batch 16 --gpus 0,1 --lr 5e-4 --load_model ../exp/ctdet_coco_hg.pth --num_workers 8 --num_epochs 320 --lr_step 270,300 2.2 test CHP_DLA_34: python test.py multi_pose --exp_id dla_USnavy512_RGuass1_12raduis_arf --arch dlaarf_34 --dataset coco_hp --resume

eval a model

1. Test the model to get a result.json file.
2. Make DOTA format ground-truth file by ./src/4rotate_xml2DOTAtxt.py
   Here, we give an example for DOTA format ground-truth file in ./Figs/GT20_TxT
3. run to eval on test set by run ./srv/5eval_json.py

Citation

If you find this project useful for your research, please use the following BibTeX entry.

@article{CHPDet, title={Arbitrary-Oriented Ship Detection through Center-Head Point Extraction}, author={Zhang, Feng and Wang, Xueying and Zhou, Shilin and Wang, Yingqian and Hou, Yi}, journal={IEEE Transactions on Geoscience and Remote Sensing}, year={2021}, publisher={IEEE} }

Contact

Any question regarding this work can be addressed to zhangfeng01@nudt.edu.cn.