MaizeMask: Low-Cost UAV RGB Dataset for Hybrid Weed-Region and Visible Maize Instance Segmentation

MaizeMask v1.0 is a curated low-cost UAV RGB dataset for hybrid field-image segmentation. It contains 332 anonymized 640 x 640 PNG tiles and 12,890 COCO segmentation annotations from three anonymized fields.The release supports semantic maize/weed segmentation and visible-maize instance segmentation. Its five source labels are maize-u, maize2, maize4, maize6, and weed. Weed annotations represent semantic cover regions, not individual plants; maize2/4/6 are appearance-based visible groups, not biological V-stages. The primary evaluation protocol is leave-one-field-out; a spatially guarded standard split and a leakage-prone random-tile control are included.The package excludes raw captures, exact coordinates, capture timestamps, source paths, raw video, and internal audits. See README.md and annotations/task_definitions.json for task definitions, split guidance, and usage constraints.