Skip to content

YOLOv5 in DOTA with CSL_label.(Oriented Object Detection)(Rotation Detection)(Rotated BBox)基于YOLOv5的旋转目标检测

License

Notifications You must be signed in to change notification settings

Tomingz/YOLOv5_DOTA_OBB

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

48 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

YOLOv5_DOTA_OBB

YOLOv5 in DOTA_OBB dataset with CSL_label.(Oriented Object Detection)

Datasets and pretrained checkpoint

Fuction

  • train.py. Train.

  • detect.py. Detect and visualize the detection result. Get the detection result txt.

  • evaluation.py. Merge the detection result and visualize it. Finally evaluate the detector

Installation (Linux Recommend, Windows not Recommend)

1. Python 3.8 with all requirements.txt dependencies installed, including torch==1.6, opencv-python==4.1.2.30, To install run:

$   pip install -r requirements.txt

2. Install swig

$   cd  \.....\yolov5_DOTA_OBB\utils
$   sudo apt-get install swig

3. Create the c++ extension for python

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

More detailed explanation

想要了解相关实现的细节和原理可以看我的知乎文章:

Usage Example

1. 'Get Dataset'

  • Split the DOTA_OBB image and labels. Trans DOTA format to YOLO longside format.

  • You can refer to hukaixuan19970627/DOTA_devkit_YOLO.

  • The Oriented YOLO Longside Format is:

$  classid    x_c   y_c   longside   shortside    Θ    Θ∈[0, 180)


* longside: The longest side of the oriented rectangle.

* shortside: The other side of the oriented rectangle.

* Θ: The angle between the longside and the x-axis(The x-axis rotates clockwise).x轴顺时针旋转遇到最长边所经过的角度

WARNING: IMAGE SIZE MUST MEETS 'HEIGHT = WIDTH'

label_format_demo

2. 'train.py'

  • All same as ultralytics/yolov5. You better train demo files first before train your custom dataset.
  • Single GPU training:
$ python train.py  --batch-size 4 --device 0
  • Multi GPU training: DistributedDataParallel Mode
python -m torch.distributed.launch --nproc_per_node 4 train.py --sync-bn --device 0,1,2,3

train_batch_mosaic0 train_batch_mosaic1 train_batch_mosaic2

3. 'detect.py'

  • Download the demo files.
  • Then run the demo. Visualize the detection result and get the result txt files.
$  python detect.py

detection_result_before_merge1 detection_result_before_merge2 draw_detection_result

4. 'evaluation.py'

  • Run the detect.py demo first. Then change the path with yours:
evaluation
(
        detoutput=r'/....../DOTA_demo_view/detection',
        imageset=r'/....../DOTA_demo_view/row_images',
        annopath=r'/....../DOTA_demo_view/row_DOTA_labels/{:s}.txt'
)
draw_DOTA_image
(
        imgsrcpath=r'/...../DOTA_demo_view/row_images',
        imglabelspath=r'/....../DOTA_demo_view/detection/result_txt/result_merged',
        dstpath=r'/....../DOTA_demo_view/detection/merged_drawed'
)
  • Run the evaluation.py demo. Get the evaluation result and visualize the detection result which after merged.
$  python evaluation.py

detection_result_after_merge

有问题反馈

在使用中有任何问题,欢迎反馈给我,可以用以下联系方式跟我交流

  • 知乎(@略略略
  • 代码问题提issues,其他问题请知乎上联系

感激

感谢以下的项目,排名不分先后

关于作者

  Name  : "胡凯旋"
  describe myself:"咸鱼一枚"
  

更多

  • 大家也可以参考另一个项目BossZard/rotation-yolov5. , 该项目在比较早的时候就将yolov5用于旋转目标检测,也是本项目的启发之一;
  • 更多的旋转目标检测器可以参考SJTU-Thinklab-Det/DOTA-DOAI. 里面有配套论文与开源代码.

About

YOLOv5 in DOTA with CSL_label.(Oriented Object Detection)(Rotation Detection)(Rotated BBox)基于YOLOv5的旋转目标检测

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 51.7%
  • C++ 42.9%
  • Cuda 3.7%
  • Shell 1.1%
  • Dockerfile 0.4%
  • Cython 0.2%