fMoW stands for Functional Map of the World. It is an image dataset used to develop and train machine learning models that can predict the function and purpose of buildings and land use based on the time series of satellite images and a large number of metadata features. The dataset contains more than 1 million images from more than 200 countries, with 63 categories of annotations, and each image has at least 1 category of annotation.

There are two versions of this dataset, fMoW-full and fMoW-rgb. fMoW-full is in TIFF format, contains 4-band and 8-band multispectral images, and has a large capacity of about 3.5 TB. fMoW-rgb is in JPEG format, and all multispectral images have been converted to RGB format. It is significantly smaller, about 200 GB.