We are proud to present the International Workshop on Computer Vision for UAVs! This year, the UAVision workshop organizers teamed up with VisDrone in a combined, full day workshop. This workshop is in conjunction with ECCV2020, SEC, Glasgow and is scheduled on August 28th.

This workshop focuses on two different trends concerning Computer Vision for UAVs, and consists of a regular paper track (accepted papers will be presented as posters), and five different challenges.

Firstly (UAVision), from the viewpoint of state-of-the-art real-time image processing on-board of Unmanned Aerial Vehicles. Indeed, cameras make ideal sensors for drones as they are lightweight, power-efficient and an enormously rich source of information about the environment in numerous applications. Although lots of information can be derived from camera images using the newest computer vision algorithms, the use of them on-board of UAVs poses unique challenges. Their computational complexity often conflicts with the need for real-time operation and the extreme resource limitations of the platform. Of course, developers have the choice to run their image processing on-board or on a remote processing device, although the latter requires a wireless link with high bandwidth, minimal latency and ultra-reliable connection. Indeed, truly autonomous drones should not have to rely on a wireless datalink, thus on-board real-time processing is a necessity. However, because of the limitations of UAVs (lightweight processing devices, limited on-board computational power, limited electrical power on-board), extreme algorithmic optimization and deployment on state-of-the-art embedded hardware (such as embedded GPUs) is the only solution.

Secondly (VisDrone), drone-based vision has drawn increasing attention in recent years with many applications, including aerial photography, surveillance, etc. Despite great progress in general computer vision algorithms, these algorithms are usually not optimal on drone platforms. Thus, developing and evaluating new vision algorithms for drone generated visual data becomes a key problem in drone-based applications. However, this goal is heavily limited by the lack of large-scale dedicated benchmarks and assessment of vision algorithms on drone platform.

Apart from the regular paper submissions track we also organize a challenge with five different tasks.