Label360

An annotation interface for labelling instance-aware semantic labels on panoramic full images, which outperforms other annotation tools in terms of efficiency and quality. Here is our demo site (https://label-360.droneye.tw).

Dataset

We also provides high quality dense annotations. If you find the dataset is useful, please cite us.

Polygon Annotations

Download Dataset

• Total # of images: 370
• Scenes
  • NTU: 3
  • Tainan street view: 286
  • Tai-61 express way: 81
• Total # of annotations: 72763
  • Building: 33316
  • Buildings: 7040
  • BuildingIgnore: 1335
  • Bicycle: 18
  • Bicycles: 15
  • Car: 23847
  • Cars: 1201
  • CarIgnore: 113
  • FreightCar: 171
  • Motorcycle: 3940
  • Motorcycles: 96
  • Trailer: 63
  • Truck: 1363
  • Van: 80
• Total # of vertices: 363815

Bounding Box Annotations

Download Dataset

• Total # of images: 26
• Scene: NTU
• Total # of annotations: 5763
  • Bicycle: 288
  • Bicycles: 131
  • Building: 2733
  • Buildings: 208
  • BuildingIgnore: 141
  • Car: 1515
  • Cars: 35
  • CarIgnore: 3
  • FreightCar: 15
  • Motorcycle: 629
  • Motorcycles: 23
  • Truck: 38
  • Van: 4

Experiment Result

• Total # of images: 20
• Scenes:
  • NTU :5)
  • Tainan street view (5)
  • Tai-61 express way (10)
• Categories
  • Drone
  • Sky
  • Road
  • Vehicle
  • Construction
  • Nature
  • Human

Citation

@incollection{lin2020label360,
  title={Label360: An Annotation Interface for Labeling Instance-Aware Semantic Labels on Panoramic Full Images},
  author={Lin, Liang-Han and Wen, Hao-Kai and Kao, Man-Hsin and Chen, Evelyn and Lin, Tse-Han and Ouhyoung, Ming},
  booktitle={SIGGRAPH Asia 2020 Posters},
  pages={1--2},
  year={2020}
}

Acknowledgements

This project was funded by Ministry of Science and Technology (MOST).