DRASHTI-HaOBB: Drone nadiR-view Annotated imageS of veHicles dataseT for India - Heading-angle Oriented Bounding Box: Indian Vehicle Oriented-Object-Detection Dataset with 1.3 Million Samples

DRASHTI-HaOBB (Drone nadiR-view Annotated imageS of veHicles dataseT for India - Heading-angle Oriented Bounding Box): Indian Vehicle Oriented-Object-Detection Dataset with 1.3 Million Vehicle Samples Data collectors & curators: Yagnik Bhavsar, Mazad Zaveri, Mehul Raval, Shaheriar Zaveri Annotations: Oriented Bounding Box (OBB) and Heading-angle for Vehicles; Flight-height View: Drone-based Aerial-view (90° downward gimbal) Location: Ahmedabad, Gujarat, India No. of Vehicle Classes: 14 Class Labels: ‘Auto3WCargo’, ‘AutoRicksaw’, ‘Bus’, ‘Container’, ‘Mixer’, ‘MotorCycle’, ‘PickUp’, ‘SUV’, ‘Sedan’, ‘Tanker’, ‘Tipper’, ‘Trailer’, ‘Truck’, ‘Van’ No. of total Images: 27,577 (84.73% are real-world images and 15.27% are augmented* images) No. of total (vehicle) Samples: 1,308,989 Dimension of each 4K image: 3840x2160 AI - Computer Vision Task: Oriented Object Detection - (Compatible with Ultralytics YOLO and OpenMMLab MMRotate) Classwise distribution of the dataset: Split Train Val Test Class Real Augmented Real Augmented Real Augmented Auto3WCargo 8652 9888 4394 4934 1519 1732 AutoRicksaw 56836 14279 28608 7189 9543 2522 Bus 1363 13864 666 7106 260 2045 Container 7429 9397 3766 4576 1179 1791 Mixer 1752 4355 829 2217 285 740 MotorCycle 323856 3115 162306 1492 53557 592 PickUp 19284 7147 9497 3610 3167 1238 SUV 164457 1293 82274 676 27042 248 Sedan 46188 810 23058 423 7605 155 Tanker 5900 6014 2930 2981 966 1145 Tipper 9576 4043 4690 2014 1555 734 Trailer 6116 8026 3075 3880 1005 1423 Truck 25240 11851 12657 5703 4261 1915 Van 10644 4015 5288 2072 1750 714 *Data augmentation uses "copy-paste" technique (Ghiasi et al., CVPR 2021), where additional samples of minority vehicle classes are pasted on real-world images containing parked vehicles, to mitigate the class imbalance problem.