EO by Day, IR by Night: Public Paired EO-IR Pixel-Level Anti-UAV Segmentation Benchmark (masks, motion-state labels, splits, and code)

SAM2-derived binary UAV segmentation masks for BOTH streams (visible/EO and thermal-infrared/IR) of the public Anti-UAV tracking dataset, together with: per-frame motion-state labels (hover/slow/fast, derived from the source box trajectories), recording-disjoint train/val/test split manifests, 5-fold recording-level cross-validation folds, CST Anti-UAV transfer-protocol splits, and the full evaluation, quality-control, and mask-generation code. Licensing: derived masks and labels are CC BY-NC 4.0; code is MIT (see README.md inside the archive). The source image frames are NOT re-hosted here — they remain with the original Anti-UAV distributors under their own terms; an end-to-end regeneration pipeline (frame download, box-prompted SAM2 mask generation, split construction) is included so the complete benchmark can be rebuilt from the public source. Companion manuscript: "EO by Day, IR by Night: A Public Paired EO-IR Pixel-Level Anti-UAV Segmentation Benchmark and Hard-Tail Study" (under review, IEEE Transactions on Multimedia). A Gebru-style datasheet (DATASHEET.md) documents collection, annotation, intended use, and limitations. This work was supported by the Korea Research Institute for Defense Technology Planning and Advancement (KRIT) grant funded by the Korea government (DAPA) (KRIT-CT-23-041).