Forest road crack dataset

synchronized acquisition was conducted in representative mountainous road regions in Hunan, China, including the Huifeng Forest Farm in Yongzhou, using an in vehicle high definition camera system and a low altitude UAV; the collection covers multiple road structures such as gravel roads and concrete pavements, and includes sunny, overcast, and post rainfall wet conditions so as to preserve typical disturbances such as cast shadows and water stains. The original video frames and still images were then manually curated by removing motion blurred and redundant frames, applying unified cropping to focus on crack regions, and standardizing the resolution to 512 × 512, yielding 1100 high quality crack images; after per image verification, LabelImg was used to produce precise bounding box annotations for crack regions.