YOLOv5 for DOTAv1.5 datasets

---

Orginal project site:

1. DOTA dataset:https://captain-whu.github.io/DOTA/dataset.html
2. YOLOv5:https://github.com/ultralytics/yolov5

Feature

Training models for detecting ships in aerial images.

This project can extract images containing ships that its gsd is greater than 0.1 and less than 0.5 in the DOTA dataset, and crop these aerial images into 800*800 pieces for training.

The recognition effect is as shown above.

How to train your own model

---

0. install dependencies

pip install -U -r requirements.txt

1. Put training images into .\data\src\img,label txt into .\data\src\otxt
2. Run python caijian.py to crop images into pieces
3. Run python train_crop.py to select images that are ships(or others)
4. Run first.py and voc_label.py to Convert dota label format to yolo label format
5. Modify configs in .\data\icon.yaml and .\models\yolov5*.yaml
6. Run to train your own model

python train.py --data icon.yaml --cfg  yolov5*.yaml --weights yolov5*.pt --batch-size 16

7. Use connect_img.py to connect cropped images NOTICE: Remember to modify the path in some files.

---

Inference

Run

python detect.py --source ./inference/images/ --weights best.pt --conf 0.4

Others

You can also train a models to recognize the rotation of these ships by using these project below:

1. R2CNN:https://github.com/yangxue0827/R2CNN_FPN_Tensorflow
2. YOLT:https://github.com/CosmiQ/yolt