NAT2021 Benchmark

NAT2021 gives a comprehensive performance evaluation of nighttime aerial tracking and provides adequate unlabelled nighttime tracking videos for unsupervised training.

Images in NAT2021 are captured in diverse nighttime scenes (e.g., roads, urban landscapes, and campus) by a DJI Mavic Air 2 UAV in a frame rate of 30 frames/s. Sequence categories consist of a wide variety of targets (e.g., cars, trucks, persons, groups, buses, buildings, and motorcycles) or activities (e.g., cycling, skating, running, and ball games)

  • NAT2021-test: 180 sequences, 141k frames
  • NAT2021-L-test: 23 sequences, 54k frames
  • NAT2021-train: 1400 sequences, 276k frames
Paper Github


The vision4robotics group is a multidisciplinary research group at Tongji University. Our research interests focus on intelligent vision and control technologies for robotics.

Get In Touch

Kaiwu Building, Tongji University, No.4800 Cao’an Highway, Jiading District, Shanghai #201804, China