Datasets:

aegean-ai
/

dut-anti-uav

image_id stringlengths 5 5	width int32 240 5.62k	height int32 160 3.74k	objects dict
00001	550	412	{ "bbox": [ [ 228, 155, 353, 245 ] ], "category": [ 0 ] }
00002	1,000	667	{ "bbox": [ [ 272, 107, 765, 491 ] ], "category": [ 0 ] }
00003	511	317	{ "bbox": [ [ 100, 67, 417, 271 ] ], "category": [ 0 ] }
00004	479	261	{ "bbox": [ [ 105, 50, 381, 216 ] ], "category": [ 0 ] }
00005	579	388	{ "bbox": [ [ 120, 160, 460, 357 ] ], "category": [ 0 ] }
00006	451	258	{ "bbox": [ [ 178, 101, 280, 150 ] ], "category": [ 0 ] }
00007	477	246	{ "bbox": [ [ 92, 56, 375, 176 ] ], "category": [ 0 ] }
00008	600	428	{ "bbox": [ [ 179, 121, 288, 195 ] ], "category": [ 0 ] }
00009	600	422	{ "bbox": [ [ 248, 100, 347, 153 ] ], "category": [ 0 ] }
00010	500	298	{ "bbox": [ [ 196, 9, 489, 180 ] ], "category": [ 0 ] }
00011	560	412	{ "bbox": [ [ 121, 108, 448, 267 ] ], "category": [ 0 ] }
00012	478	280	{ "bbox": [ [ 93, 60, 374, 180 ] ], "category": [ 0 ] }
00013	600	420	{ "bbox": [ [ 114, 108, 473, 312 ] ], "category": [ 0 ] }
00014	479	261	{ "bbox": [ [ 104, 48, 380, 212 ] ], "category": [ 0 ] }
00015	660	330	{ "bbox": [ [ 272, 30, 582, 233 ] ], "category": [ 0 ] }
00016	645	323	{ "bbox": [ [ 154, 53, 254, 128 ] ], "category": [ 0 ] }
00017	240	160	{ "bbox": [ [ 108, 41, 172, 94 ] ], "category": [ 0 ] }
00018	940	626	{ "bbox": [ [ 350, 245, 590, 365 ] ], "category": [ 0 ] }
00019	750	500	{ "bbox": [ [ 125, 93, 243, 190 ] ], "category": [ 0 ] }
00020	1,000	667	{ "bbox": [ [ 392, 259, 605, 382 ] ], "category": [ 0 ] }
00021	1,790	1,080	{ "bbox": [ [ 654, 451, 778, 541 ], [ 946, 212, 1134, 345 ] ], "category": [ 0, 0 ] }
00022	800	534	{ "bbox": [ [ 321, 61, 704, 306 ] ], "category": [ 0 ] }
00023	590	350	{ "bbox": [ [ 187, 75, 447, 275 ] ], "category": [ 0 ] }
00024	960	635	{ "bbox": [ [ 417, 249, 579, 415 ] ], "category": [ 0 ] }
00025	660	398	{ "bbox": [ [ 167, 102, 534, 326 ] ], "category": [ 0 ] }
00026	634	423	{ "bbox": [ [ 154, 89, 500, 343 ] ], "category": [ 0 ] }
00027	640	480	{ "bbox": [ [ 105, 73, 443, 304 ] ], "category": [ 0 ] }
00028	800	601	{ "bbox": [ [ 178, 170, 494, 400 ] ], "category": [ 0 ] }
00029	852	480	{ "bbox": [ [ 318, 111, 474, 255 ] ], "category": [ 0 ] }
00030	852	480	{ "bbox": [ [ 548, 161, 721, 271 ] ], "category": [ 0 ] }
00031	800	450	{ "bbox": [ [ 225, 87, 572, 414 ] ], "category": [ 0 ] }
00032	634	420	{ "bbox": [ [ 136, 79, 498, 360 ] ], "category": [ 0 ] }
00033	350	350	{ "bbox": [ [ 46, 65, 291, 263 ] ], "category": [ 0 ] }
00034	776	517	{ "bbox": [ [ 227, 17, 299, 71 ] ], "category": [ 0 ] }
00035	600	404	{ "bbox": [ [ 234, 155, 533, 265 ] ], "category": [ 0 ] }
00036	720	405	{ "bbox": [ [ 192, 55, 426, 215 ] ], "category": [ 0 ] }
00037	630	420	{ "bbox": [ [ 152, 81, 422, 283 ] ], "category": [ 0 ] }
00038	634	420	{ "bbox": [ [ 112, 93, 565, 243 ] ], "category": [ 0 ] }
00039	800	400	{ "bbox": [ [ 439, 108, 643, 294 ] ], "category": [ 0 ] }
00040	923	498	{ "bbox": [ [ 240, 74, 695, 418 ] ], "category": [ 0 ] }
00041	1,280	853	{ "bbox": [ [ 297, 196, 1009, 497 ] ], "category": [ 0 ] }
00042	750	493	{ "bbox": [ [ 113, 115, 360, 245 ] ], "category": [ 0 ] }
00043	610	380	{ "bbox": [ [ 251, 19, 335, 57 ] ], "category": [ 0 ] }
00044	540	458	{ "bbox": [ [ 184, 125, 348, 253 ] ], "category": [ 0 ] }
00045	640	320	{ "bbox": [ [ 138, 31, 412, 272 ] ], "category": [ 0 ] }
00046	600	450	{ "bbox": [ [ 131, 121, 473, 377 ] ], "category": [ 0 ] }
00047	960	540	{ "bbox": [ [ 351, 53, 862, 402 ] ], "category": [ 0 ] }
00048	580	387	{ "bbox": [ [ 124, 61, 400, 305 ] ], "category": [ 0 ] }
00049	768	432	{ "bbox": [ [ 206, 82, 503, 241 ] ], "category": [ 0 ] }
00050	980	490	{ "bbox": [ [ 298, 139, 662, 404 ] ], "category": [ 0 ] }
00051	388	292	{ "bbox": [ [ 33, 41, 315, 237 ] ], "category": [ 0 ] }
00052	852	480	{ "bbox": [ [ 204, 103, 380, 232 ] ], "category": [ 0 ] }
00053	650	434	{ "bbox": [ [ 177, 114, 465, 291 ] ], "category": [ 0 ] }
00054	600	420	{ "bbox": [ [ 157, 137, 472, 384 ] ], "category": [ 0 ] }
00055	2,146	1,413	{ "bbox": [ [ 680, 326, 1605, 862 ] ], "category": [ 0 ] }
00056	1,024	576	{ "bbox": [ [ 339, 11, 925, 475 ] ], "category": [ 0 ] }
00057	700	394	{ "bbox": [ [ 319, 99, 544, 269 ] ], "category": [ 0 ] }
00058	640	360	{ "bbox": [ [ 261, 63, 517, 287 ] ], "category": [ 0 ] }
00059	3,856	2,571	{ "bbox": [ [ 914, 532, 2742, 2211 ] ], "category": [ 0 ] }
00060	640	394	{ "bbox": [ [ 226, 89, 629, 317 ] ], "category": [ 0 ] }
00061	1,024	576	{ "bbox": [ [ 131, 83, 855, 358 ] ], "category": [ 0 ] }
00062	1,500	1,000	{ "bbox": [ [ 314, 159, 1231, 872 ] ], "category": [ 0 ] }
00063	1,000	1,000	{ "bbox": [ [ 67, 148, 906, 778 ] ], "category": [ 0 ] }
00064	640	485	{ "bbox": [ [ 144, 123, 500, 443 ] ], "category": [ 0 ] }
00065	736	491	{ "bbox": [ [ 643, 343, 719, 404 ], [ 589, 186, 685, 260 ], [ 389, 220, 507, 310 ], [ 303, 339, 419, 408 ], [ 105, 265, 229, 356 ], [ 127, 43,...
00066	1,200	704	{ "bbox": [ [ 154, 177, 856, 605 ] ], "category": [ 0 ] }
00067	768	512	{ "bbox": [ [ 161, 128, 591, 369 ] ], "category": [ 0 ] }
00068	620	349	{ "bbox": [ [ 92, 30, 534, 347 ] ], "category": [ 0 ] }
00069	1,440	612	{ "bbox": [ [ 513, 50, 645, 129 ] ], "category": [ 0 ] }
00070	2,000	1,428	{ "bbox": [ [ 391, 185, 1555, 1088 ] ], "category": [ 0 ] }
00071	1,500	1,000	{ "bbox": [ [ 346, 204, 1233, 809 ] ], "category": [ 0 ] }
00072	960	653	{ "bbox": [ [ 173, 63, 787, 581 ] ], "category": [ 0 ] }
00073	580	325	{ "bbox": [ [ 105, 80, 446, 245 ] ], "category": [ 0 ] }
00074	736	537	{ "bbox": [ [ 71, 42, 703, 481 ] ], "category": [ 0 ] }
00075	573	360	{ "bbox": [ [ 124, 64, 449, 293 ] ], "category": [ 0 ] }
00076	5,616	3,744	{ "bbox": [ [ 1190, 747, 4556, 2891 ] ], "category": [ 0 ] }
00077	1,280	720	{ "bbox": [ [ 293, 53, 976, 559 ] ], "category": [ 0 ] }
00078	1,920	1,080	{ "bbox": [ [ 165, 122, 1904, 895 ] ], "category": [ 0 ] }
00079	1,500	844	{ "bbox": [ [ 265, 123, 1240, 736 ] ], "category": [ 0 ] }
00080	1,920	1,080	{ "bbox": [ [ 296, 149, 1178, 831 ] ], "category": [ 0 ] }
00081	615	409	{ "bbox": [ [ 118, 63, 517, 299 ] ], "category": [ 0 ] }
00082	1,000	519	{ "bbox": [ [ 233, 139, 817, 506 ] ], "category": [ 0 ] }
00083	660	371	{ "bbox": [ [ 173, 59, 539, 328 ] ], "category": [ 0 ] }
00084	1,200	742	{ "bbox": [ [ 155, 124, 709, 583 ] ], "category": [ 0 ] }
00085	530	298	{ "bbox": [ [ 102, 9, 460, 265 ] ], "category": [ 0 ] }
00086	471	284	{ "bbox": [ [ 131, 26, 404, 203 ] ], "category": [ 0 ] }
00087	800	450	{ "bbox": [ [ 277, 120, 537, 355 ] ], "category": [ 0 ] }
00088	815	360	{ "bbox": [ [ 151, 21, 670, 292 ] ], "category": [ 0 ] }
00089	3,096	2,064	{ "bbox": [ [ 2054, 775, 2844, 1287 ], [ 272, 691, 970, 1005 ] ], "category": [ 0, 0 ] }
00090	1,200	630	{ "bbox": [ [ 407, 52, 893, 444 ] ], "category": [ 0 ] }
00091	660	371	{ "bbox": [ [ 299, 62, 502, 249 ] ], "category": [ 0 ] }
00092	1,920	1,080	{ "bbox": [ [ 397, 177, 1437, 887 ] ], "category": [ 0 ] }
00093	1,050	550	{ "bbox": [ [ 119, 83, 800, 452 ] ], "category": [ 0 ] }
00094	620	300	{ "bbox": [ [ 217, 35, 538, 249 ] ], "category": [ 0 ] }
00095	1,268	760	{ "bbox": [ [ 297, 94, 1057, 586 ] ], "category": [ 0 ] }
00096	407	418	{ "bbox": [ [ 33, 79, 362, 321 ] ], "category": [ 0 ] }
00097	392	392	{ "bbox": [ [ 85, 38, 184, 93 ] ], "category": [ 0 ] }
00098	660	371	{ "bbox": [ [ 227, 40, 612, 304 ] ], "category": [ 0 ] }
00099	1,024	640	{ "bbox": [ [ 410, 12, 960, 379 ] ], "category": [ 0 ] }
00100	736	412	{ "bbox": [ [ 128, 59, 504, 275 ] ], "category": [ 0 ] }

End of preview. Expand in Data Studio

DUT Anti-UAV

This is a redistribution of the DUT Anti-UAV dataset, created by Jie Zhao, Jingshu Zhang, Dongdong Li and Dong Wang at Dalian University of Technology. It holds both parts of the dataset, detection and tracking. It is uploaded here so that students and researchers can download it with the Hugging Face tools. All credit for the data belongs to the original authors.

Original repository: https://github.com/wangdongdut/DUT-Anti-UAV
Paper: https://arxiv.org/abs/2205.10851
License: Apache-2.0, the license the authors attach to the repository. A copy is in LICENSE.

Every zip file here is byte-for-byte the file the authors published on Google Drive. Nothing was converted, renamed or removed. The SHA-256 of each file is listed below so you can check your copy.

Detection

10,000 visible-light images of drones, split into train, validation and test sets. Each image has a Pascal VOC XML annotation with one or more boxes of the class UAV.

File	Images	SHA-256
`detection/train.zip`	5,200	`14f927290556df60e23cedfa80dffc10dc21e4a3b6843e150cfc49644376eece`
`detection/val.zip`	2,600	`238be0ceb3e7c5be6711ee3247e49df2750d52f91f54f5366c68bebac112ebf8`
`detection/test.zip`	2,200	`a671989a01cff98c684aeb084e59b86f4152c50499d86152eb970a9fc7fb1cbe`

Each zip unpacks to <split>/img/NNNNN.jpg and <split>/xml/NNNNN.xml.

Parquet copy of the detection set

detection-parquet/ holds the same detection images and boxes as Parquet files, so you can browse them in the dataset viewer and load them with the datasets library. This copy is derived from the zip files above: each row is one image, stored unchanged as JPEG bytes, with the boxes read from its XML file. Nothing else was changed. The three splits have 5,243, 2,621 and 2,245 boxes; three training images have none.

Column	Contents
`image_id`	the file name without `.jpg`, as in the zip
`image`	the image
`width`, `height`	image size in pixels, from the XML file
`objects.bbox`	one box per drone as `[xmin, ymin, xmax, ymax]` in pixels, the Pascal VOC corners
`objects.category`	always `UAV`

from datasets import load_dataset

ds = load_dataset("aegean-ai/dut-anti-uav", "detection")

Tracking

20 visible-light videos of a single drone, with short-term and long-term sequences, stored as 24,804 JPEG frames. Every frame has a ground-truth line.

File	Contents	SHA-256
`tracking/Anti-UAV-Tracking-V0.zip`	frames, one directory per video (`video01` to `video20`)	`5f3bb7150d38e613243397555405258674776b795b90b1e8c98c5259fe1bd72f`
`tracking/Anti-UAV-Tracking-V0GT.zip`	ground truth, one file per video (`videoNN_gt.txt`)	`8722a642e2f0e0981faf728efff4bd19e0e0e1b65df723e0a4ed1f11c972648f`

Each line of a ground-truth file describes one frame, in frame order. It holds four integers: the x and y pixel coordinates of the box's top-left corner, then its width and height in pixels. A frame where the drone is not visible is marked -100 -100 -100 -100; 2,586 frames in 9 of the 20 videos carry that marker.

Each video directory in the frames zip also holds videoNN_gt_first.txt, which repeats the first frame's box. Single-object trackers start from that box. Every video's ground-truth file has exactly one line per frame.

Video	Frames	Resolution
video01	1,050	1920x1080
video02	83	1280x720
video03	100	1280x720
video04	341	1920x1080
video05	450	1280x720
video06	200	1920x1080
video07	2,480	1280x720
video08	2,305	1280x720
video09	2,500	1920x1080
video10	2,635	1920x1080
video11	1,000	1920x1080
video12	1,485	1920x1080
video13	1,915	1920x1080
video14	590	1920x1080
video15	1,350	1920x1080
video16	1,285	1920x1080
video17	780	1920x1080
video18	1,320	1920x1080
video19	1,300	1920x1080
video20	1,635	1920x1080

Parquet copy of the tracking set

tracking-parquet/ holds the same frames and boxes as one Parquet file per video, so you can browse them in the dataset viewer and load them with the datasets library. This copy is derived from the zip files above: each row is one frame, stored unchanged as JPEG bytes, with its box from the ground-truth file. The config has a single split, test, because the paper evaluates trackers on all 20 videos.

Column	Contents
`video_id`	`video01` to `video20`
`frame_index`	frame number, counted from 1, as in the frame file names
`image`	the frame
`width`, `height`	frame size in pixels
`bbox`	the ground-truth box as `[x, y, width, height]` in pixels, from the top-left corner, or `[-100, -100, -100, -100]` when the drone is not visible

from datasets import load_dataset

frames = load_dataset("aegean-ai/dut-anti-uav", "tracking", split="test")
video05 = frames.filter(lambda v: v == "video05", input_columns="video_id")

Preview videos

tracking-preview/videoNN.mp4 is one H.264 video per sequence for watching the data, with the ground-truth box drawn in green on frames where the drone is visible. The previews are derived and lossy: frames are scaled to at most 1280 pixels wide and re-encoded. The dataset does not state a frame rate, so the previews play at 30 frames per second. Use the frames or the Parquet copy, not the previews, for evaluation.

Download

hf download aegean-ai/dut-anti-uav --repo-type dataset --local-dir dut-anti-uav

Citation

If you use this data, cite the original paper:

@article{zhao2022dutantiuav,
  title   = {Vision-based Anti-UAV Detection and Tracking},
  author  = {Zhao, Jie and Zhang, Jingshu and Li, Dongdong and Wang, Dong},
  journal = {IEEE Transactions on Intelligent Transportation Systems},
  year    = {2022}
}

Downloads last month: 1,261

Paper for aegean-ai/dut-anti-uav

Vision-based Anti-UAV Detection and Tracking

Paper • 2205.10851 • Published May 22, 2022