Drone-Bench: Detect
Agents · 2026-07-23
Drone-Bench's Detect task: given a reference photo of a person's face, the agent builds a detector that returns a whole-body bounding box around that person in each frame of real drone video, and may keep state across frames. Submissions are scored against Andon's own labels on how well the box matches while the target is on screen and on whether the detector reports the target as absent once they leave the frame. The score is the mean over ten runs of each run's best submission, in percent; Andon's human-plus-AI baseline is 72.1.
Benchmark source: Andonlabs