Dataset Hub

STAR is the first large‑scale dataset for scene‑graph generation (SGG) in high‑resolution satellite images. It contains over 210,000 objects and more than 400,000 triples across 1,273 complex global scenes.

95‑Cloud is an extension of the 38‑Cloud dataset for binary cloud segmentation in satellite imagery. It contains 34,701 training patches of size 384 × 384 derived from 75 Landsat 8 Collection 1 Level‑1 scenes, primarily located in North America. The test set is identical to that of 38‑Cloud and contains 9,201 patches. Each patch includes four spectral channels: red, green, blue, and near‑infrared.