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.
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 |
conda create -n rps python=3.11
conda activate rps
pip install -r requirements.txt
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)