M2S2: A Publicly Available Multispectral Dataset for Satellite Material Segmentation

NOTE: The data files for this dataset are currently undergoing a formal Public Release (PA) review. The full dataset (~33GB) will be uploaded to this repository upon final approval. This repository serves as the permanent citable location for the dataset.

Dataset Description

We introduce M2S2, a publicly available multispectral dataset for satellite material segmentation, comprising RGB-approximated synthetic multispectral imagery and hardware-in-the-loop (HIL) true multispectral data across three satellite geometries (Voyager, ISS, Hubble) with up to 11 material classes.

This dataset provides a standardized benchmark for developing and evaluating multispectral segmentation methods tailored to space-based scenarios. The synthetic domain consists of RGB-approximated multispectral renderings, while the HIL domain contains true multispectral imagery from the TOUCAN sensor.

Citation Information

This dataset is the official supplement to our paper: M2S2: Multispectral Dataset for Material Segmentation of Satellites

To be presented at the 15th IEEE Workshop on Hyperspectral Imaging and Signal Processing: Evolution in Remote Sensing (WHISPERS) 2025.