AERO-DETR Runway Dataset: A Georeferenced Benchmark for Runway Detection and Marking Recognition in High-Resolution Aerial Imagery
收藏资源简介:
The AERO-DETR Runway Dataset is a georeferenced benchmark for runway detection and marking recognition from sub-meter aerial imagery. It covers 125 U.S. airports across five runway configuration categories (single, parallel, intersection, mixed, complex) with imagery from public-domain NAIP and NOAA archives at 0.42–0.72 m GSD. The dataset provides dual-level annotations: 1,038 oriented bounding boxes (OBBs) delineating individual runways, and 12,816 fine-grained marking polygons across eight classes aligned with RTCA DO-272 and FAA AC 150/5340-1M. Annotations are provided in LabelMe JSON format across 433 full-scene images and 995 orientation-normalized runway crops. The deposit includes georeferenced imagery (JPEG 2000), vector annotations, per-airport metadata, and reproducible Python pipelines for dataset creation and format conversion. See README.md for full documentation. v3 update: This version now contains both the full dataset and the model checkpoints used for reproduction (models.zip): the Stage-1 oriented-bounding-box detector rwy_obb_v1.pt (MD5 CC80EBDDE3A36135E19F27F49AB2E270) and the Stage-3 RT-DETR runway-marking detector rwy_markings_H_v1.pt (MD5 9A7B748F401547E939937D98CA8C5C87).



