YOLOv10-OBB (Oriented Bounding Box)
As this repo is launched on 10th July, there is still no pretrained model of YOLOv10 that supports OBB, therefore i initiate this repo and perform the training and testing based on DOTAv1 dataset.
UPDATES
18/07/2024
- converter annotated label studio to YOLO-OBB is done. check the repo
14/07/2024
- added pretrained for Yolov10S(1024), Yolov10B, Yolov10L, Yolov10M, Yolov10X.
- wip convertion from Label studio annotation to support training for YOLO-OBB
10/07/2024
- added pretrained for Yolov10N and Yolov10S (640)
FEATURES
- [x] Pretrained YOLOv10-N-OBB
- [x] Pretrained YOLOv10-S-OBB
- [x] Pretrained YOLOv10-M-OBB
- [x] Pretrained YOLOv10-B-OBB
- [x] Pretrained YOLOv10-L-OBB
- [x] Pretrained YOLOv10-X-OBB
- [x] Hugging Face Space (Note: this will regularly updated once the trained model is finished)
PERFORMANCE
| model | size (pixels) | mAP50 test | mAP50-95 test | Link | Notes |
|---|---|---|---|---|---|
| YOLOv10-N-OBB | 1024 | 0.36 | 0.235 | Download | |
| YOLOv10-S-OBB | 640 | 0.394 | 0.256 | Download | Coincidentally, this was trained with image size 640 |
| YOLOv10-S-OBB | 1024 | 0.457 | 0.312 | Download | |
| YOLOv10-M-OBB | 1024 | 0.496 | 0.345 | Download | |
| YOLOv10-B-OBB | 1024 | 0.486 | 0.335 | Download | |
| YOLOv10-L-OBB | 1024 | 0.52 | 0.369 | Download | |
| YOLOv10-X-OBB | 1024 | 0.527 | 0.376 | Download |
INSTALATION
conda create -n rps python=3.11
conda activate rps
pip install -r requirements.txtHOW TO USE?
from ultralytics import YOLO
model = YOLO("/path/to/model.pt", task="obb")
model.predict('/path/to/test.jpg', save=True, imgsz=1024, conf=0.5)