WebUAV-3M: A Benchmark for Unveiling the Power of Million-Scale Deep UAV Tracking

• DOI: 10.48550/arxiv.2201.07425
• Publication Date: 2022-01-01
• Authors: Chunhui Zhang; Guanjie Huang; Li Liu; Shan Huang; Yinan Yang; Yuxuan Zhang; Xiang Wan; Shiming Ge
• Keywords: BitTorrent tracker; Benchmark (surveying); Tracking (education)

Unmanned aerial vehicle (UAV) tracking is of great significance for a wide range applications, such as delivery and agriculture. Previous benchmarks in this area mainly focused on small-scale problems while ignoring the amounts data, types data modalities, diversities target categories scenarios, evaluation protocols involved, greatly hiding massive power deep UAV tracking. In work, we propose WebUAV-3M, largest public benchmark to date, facilitate both development trackers. WebUAV-3M contains over 3.3 million frames across 4,500 videos offers 223 highly diverse categories. Each video densely annotated with bounding boxes by an efficient scalable semiautomatic annotation (SATA) pipeline. Importantly, take advantage complementary superiority language audio, enrich innovatively providing natural specifications audio descriptions. We believe that additions will boost future research terms exploring features cues multimodal addition, fine-grained tracking-under-scenario constraint (UTUSC) protocol seven challenging scenario subtest sets are constructed enable community develop, adapt evaluate various advanced provide extensive evaluations detailed analyses 43 representative trackers envision directions field beyond. The dataset, toolkits baseline results available at \url{https://github.com/983632847/WebUAV-3M}.