validation dataset for yolov8-obb-pose cyclone-detect-model

Dataset description

  To identify the aforementioned cyclonic cloud features, the YOLOv8-obb-pose model is configured using the YOLOv8 model framework, which combines oriented object detection (obb) and keypoint detection (pose). Specifically, a branch for keypoint prediction is added to the decoupled head module of the YOLOv8-obb model. This enables the new YOLOv8-obb-pose model to simultaneously perform automatic detection of cyclone type, center position, and oriented bounding box.The dataset specifically provide the YOLOv8-obb-pose model weights to enable other researchers to replicate the model, alongside a small validation dataset for performance evaluation and to facilitate its implementation. The validation dataset comprises 1334 Vortex-Centered Infrared (VCI) images from the Nordic sea region spanning the years 2001 and 2023, with 500 cyclone-containing images of the dataset. None of these images are involved in any training process of the model.We sincerely acknowledge the providers of the YOLO framework and the authors of the YOLOv8-obb-pose model(

https://github.com/yzqxy/ultralytics-obb_segment). To advance reproducibility and facilitate further research, we hereby publicly release the model pre-trained on the Cyclone Dataset, enabling other researchers to reuse this resource for academic purposes.

Dataset structure

1 val.zip :it  contains the validation image set and validation label set. The validation image set comprises 1,374 VCI images (generation method described in the arcticle), of which 500 images exhibit cyclonic cloud features. These correspond to 500 YOLO-format labels stored in the validation label set.

2 yolov8-obb-pose_pmcs.pt : YOLOv8-obb-pose model weights pretrained in IMPMCT datasets(https://zenodo.org/records/15113263).

3 predict_example.png: an predict example for YOLOv8-obb-pose model .

4 label principle.png: show how to label cyclone in VCI images(detailed in the arcticle).

5 yolo_val.ipynb: Python code to show model's performance on this validation dataset

performance on this validation dataset

Class

Instances

BOX(P

R

mAP50

mAP50-95)

Pose(P

R

mAP50

mAP50-95)

comma

285

0.94

0.87

0.92

0.59

0.97

0.90

0.94

0.94

spiral

215

0.83

0.88

0.85

0.45

0.88

0.96

0.93

0.93

all

500

0.88

0.88

0.88

0.52

0.92

0.93

0.94

0.94