AI Computer Vision Research
LEVIRDet-159 is the largest and most comprehensive remote sensing object detection dataset to date, with 159 categories, ~2.56 million bounding boxes, and ~700k fine-grained annotations under a multi-level taxonomy. LEVIRDetNet is a universal detector that can reliably operate across sensors, resolutions, and category systems.
We construct LEVIRDet-159, the up-to-date largest and most fine-grained remote sensing object detection dataset, with 159 categories, ~2.56 million bounding boxes, and ~700k fine-grained annotations under a multi-level taxonomy. Based on LEVIRDet-159, we propose LEVIRDetNet, a scale-hierarchy-aware detection foundation model designed for universal remote sensing detection.
Without target-domain training or fine-tuning, LEVIRDetNet ranks first on all 9 benchmarks and improves the strongest competing methods by 5.02 mAP on average, while maintaining stronger precision-recall stability under practical confidence thresholds.
Read the research paper
It covers 30 common parent categories and 159 category types across global regions and diverse imaging conditions, and surpasses existing datasets in category number, image number, annotation number, object-size coverage, and geographic coverage.
Without target-domain training or fine-tuning, LEVIRDetNet ranks first on all nine benchmarks and improves the strongest competing methods by 5.02 mAP on average, while maintaining stronger precision-recall stability under practical confidence thresholds.
LEVIRDetNet ranks first on all 9 benchmarks and improves the strongest competing methods by 5.02 mAP on average. It shows strong generalization across diverse scenes, categories, resolutions, and imaging conditions without target-domain training.
LEVIRDet-159 is, to our knowledge, the up-to-date largest, the most fine-grained, and the first million-box dataset built for broad-category universal remote sensing object detection under a unified tight horizontal bounding box protocol.
Download DatasetLEVIRDet-159 is the up-to-date largest and most fine-grained remote sensing object detection dataset, with 159 categories, ~2.56 million bounding boxes, and ~700k fine-grained annotations under a multi-level taxonomy.
Download DatasetLEVIRDetNet is a scale-hierarchy-aware detection foundation model for universal remote sensing object detection. It couples online visual Ground Sampling Distance (GSD) prediction, GSD-conditioned query modulation and allocation, and a hierarchy-aware detection head for mixed-granularity remote sensing supervision.
