NASA-IBM lunar model trains on 2M image tiles

NASA and IBM released a lunar foundation model trained on roughly 2 million image tiles, including more than 1 million one-meter camera images and nearly 964,000 multispectral images at 100-meter resolution. NASA’s release says the model, machine-learning datasets, benchmark collections, and TerraTorch integration are available to researchers, while the model card links the open repository and technical report. Scientific builders should file this as a reusable data-and-fine-tuning surface, not mission-grade truth: it turns 17 years of lunar data into a starting point, much like a four-day wet-lab feedback lane.