WetVeg-2mm: An Ultra-High-Resolution UAV Orthomosaic Dataset for Fine-Grained Riparian Vegetation Semantic Segmentation

WetVeg-2mm is an ultra-high-resolution UAV orthomosaic dataset for fine-grained riparian vegetation semantic segmentation. The dataset was constructed from UAV surveys over a representative riparian section of the Qing River in Guangxi, China, with a ground sampling distance of 2 mm. The dataset contains 2054 annotated image chips of size 1024 × 1024 pixels and provides pixel-level semantic labels for 17 semantic classes in total, including 14 representative wetland plant classes, together with water, bareland, and background. This repository provides the dataset files, official split files, and original polygon annotations for reproducible research in fine-grained riparian vegetation segmentation.