DisDop Framework Enhances Open-Vocabulary Aerial Object Detection

A novel approach known as DisDop (Distillation with Domain Priors) has been introduced for detecting aerial objects in an open vocabulary. This technique tackles the issue where conventional open-vocabulary detection models, optimized for natural images, struggle with aerial photos due to differences in perspective and limited data availability. DisDop effectively extracts multi-level domain priors from remote sensing foundation models like RemoteCLIP and DINOv3, rather than depending solely on models trained on natural images. The goal is to enhance detection precision for objects beyond fixed categories, a necessity as drone usage becomes more prevalent. The findings are outlined in a paper available on arXiv (ID: 2605.24639).