Yongshan Zhu

By explicitly disentangling language into global context, spatial relations, and object attributes, ProVG achieves state-of-the-art remote sensing visual grounding, suggesting that fine-grained linguistic cues are key to unlocking performance in this domain.