UniRef-UAV (Text+Image): leaderboard
UniRef-UAV is a referring benchmark for drone imagery in which a query may name no target, one target or several, and the model must return exactly the matching set of boxes. This board uses combined text and image queries. A query counts as correct only when the predicted boxes match the annotated set one to one at an intersection over union of at least 0.5, and an empty prediction is correct only when there is no target.
Metric: Precision@(F1=1, IoU>=0.5) (%; share of the 10,322 text+image test queries whose predicted box set exactly matches the referred target set under one-to-one IoU>=0.5 matching, an empty prediction being correct only for no-target queries; zero-shot, model-specific grounding prompt, unparseable output read as an empty prediction). Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.6 27B | 58.5 | #341 |
| 2 | Gemma 4 31B (IT) | 46.2 | #342 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=uniref-uav-text-plusimage · How It Works · Data refreshed daily, snapshot 2026-11-13.