{"url":"/dataset/uav-veid","name":"UAV-VeID","full_name":null,"description_markdown":"1. Data Collection\r\nWe simulate real scenarios as much as possible during the UAV videos collection. \r\nSpecifically, UAV videos are collected from different locations with distinct backgrounds and lighting conditions, e.g., including highways, urban road intersections, parking lots, etc. \r\nFor vehicles at parking lots, we adopt various UAV sport modes such as cruising and rotating to record vehicles. \r\nThis strategy introduces viewpoint and scale changes, as well as partial occlusions to images of the same vehicle. \r\nFor moving vehicles, we use two UAVs to simultaneously shoot videos from different viewpoints and heights. \r\nThis strategy introduces viewpoint, scale, and background changes. \r\nThe flying height of UAVs ranges from 15 to 60 meters, leading to different scales of vehicle images. \r\nThe vertical angle of UAV camera ranges from 40 to 80 degrees, which leads to different viewpoints of vehicle images. \r\nThe videos are recorded at 30 frames per second (fps), with the resolution of 2704 × 1520 pixels and 4096 × 2160 pixels, respectively. \r\nThe UAV-VeID is constructed from 80 video sequences selected from raw UAV videos.\r\n\r\n2. Annotation\r\nWe annotate vehicles from collected videos to construct the UAV-VeID. \r\nIn each video clip, 1 video frame is sampled every one second to construct a video frame dataset. \r\nThe dataset annotation is hence conducted based on those sample video frames.\r\nTo finish the vehicle annotation, 6 domain experts are involved to manually locate and annotate the identities of vehicles from each video frame.\r\nThe data annotation procedure takes 1000 man-hours and finally results in a dataset containing 41,917 vehicle bounding boxes of 4601 vehicles. \r\nEach vehicle is annotated by at least two bounding boxes. \r\n\r\n3. Dataset partition\r\nThe UAV-VeID dataset is split into the training set, validation set, and testing set, among which the training set contains 18,709 images with 1,797 IDs, the validation set contains 4,150 images with 596 IDs, and the testing set contains 19,058 images with 2,208 IDs. \r\nThe validation set is further divided into a query set (\"val_q_label.txt\" 3,554 images) and a gallery set (\"val_g_label.txt\" 596 images). \r\nThe testing set is further divided into a query set (\"test_q_label.txt\" 16,850 images) and a gallery set (\"test_g_label.txt\" 2,208 images). \r\n\r\n4. Download\r\nPlease sign the Agreement(UAV-VeID_AGREEMENT.pdf) and thereby agrees to observe the restrictions listed in this document. \r\nAfter filling it, please send the electrical version to us. After confirming your information, we will send the download link and password to you via Email.\r\n\r\n5. Contact\r\nShangzhi Teng, Email: tengshangzhi@126.com","description_withheld":null,"homepage":"https://github.com/tengshangzhi/UAV-VeID","introduced_date":null,"introduced_date_note":null,"introduced_by":null,"license":null,"modalities":[],"tasks":[],"languages":[],"variants":["UAV-VeID"],"data_loaders":[{"repo":"https://github.com/tengshangzhi/UAV-VeID","url":"https://github.com/tengshangzhi/UAV-VeID/blob/main/README.md","frameworks":[]}],"num_papers_in_archive":1,"source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28"},"benchmarks":[],"papers_with_a_benchmark_row":[],"syntology_totals":{"read_at":"2026-09-24T18:15:14+00:00","papers_with_samples":0,"samples_harvested":0,"samples_ran":0,"samples_unverified":0,"pointer_only_for_licence":0,"papers_with_no_sample_that_ran":0,"note":"the per-paper counts above, summed; not a rate"},"papers_note":"The archive never published its papers-using-dataset list; these are papers with a leaderboard row on this dataset's benchmarks."}