Open-Text Aerial Detection: A Unified Framework For Aerial Visual Grounding And Detection

Open-Vocabulary Aerial Detection (OVAD) and Remote Sensing Visual Grounding (RSVG) have emerged as two key paradigms for aerial scene understanding. However, each paradigm suffers from inherent limitations when operating in isolation: OVAD is restricted to coarse category-level semantics, while RSVG is structurally limited to single-target localization. These limitations prevent existing methods from simultaneously supporting rich semantic understanding and multi-target detection. To address this, we propose OTA-Det, the first unified framework that bridges both paradigms into a cohesive architecture. Specifically, we introduce a task reformulation strategy that unifies task objectives and supervision mechanisms, enabling joint training across datasets from both paradigms with dense supervision signals. Furthermore, we propose a dense semantic alignment strategy that establishes explicit correspondence at multiple granularities, from holistic expressions to individual attributes, enabling fine-grained semantic understanding. To ensure real-time efficiency, OTA-Det builds upon the RT-DETR architecture, extending it from closed-set detection to open-text detection by introducing several high efficient modules, achieving state-of-the-art performance on six benchmarks spanning both OVAD and RSVG tasks while maintaining real-time inference at 34 FPS.

Imagine asking an aerial AI: "find all red trucks near the warehouse." It's a natural request requiring two abilities at once, handling rich queries and returning multiple matches, but no existing approach offers both. One, called Open-Vocabulary Aerial Detection (OVAD), finds every match in a scene but only accepts simple labels like "truck" or "building". The other, called Remote Sensing Visual Grounding (RSVG), interprets rich descriptions but returns only one match at a time. We wondered: could a single model handle both, and even more flexible input? We built OTA-Det by reformulating the two tasks as one, training one model that accepts any text input: a word, a sentence, or a structured list of properties (color, category, position). Surprisingly, this unified approach not only handles both modes but beats specialized models on six standard benchmarks, while satisfying real-time inference requirements. Thus, operators no longer need to pick specialized models or conform to rigid query formats. They describe what they want however feels natural, useful for everything from drone navigation to logistics and traffic monitoring. Aerial perception becomes less a tool you learn to operate, more a system that learns to understand you.