The OPS-SAT case dataset

The "OPS-SAT case" dataset is the official dataset of <strong>ESA's Kelvins</strong> <strong>"the OPS-SAT case" challenge, </strong>created in collaboration<strong> with ESA's Phi-lab </strong>and <strong>ESA's OPS-SAT spacecraft operations team</strong>. It consists of<strong> 26 raw and unprocessed images</strong> of Earth that have been taken by the OPS-SAT cube-sat using its on-board camera. The images have a resolution of around <strong>2048x1944 </strong>or similar and are provided in <strong>.png</strong> format. The goal of the competition is, given a model of a neural network (in this case the <strong>EfficientNet-Lite0</strong>), to provide best possible network parameters to perform an <strong>on-board classification</strong> task. For each of the 8 target classes, there are only 10 labelled images provided, severely limiting the number of "shots" that the neural network has on learning. The size of the labeled patches is <strong>200x200</strong> pixels and the labels are related to the following types of landcover/content: <strong>Agricultural, Cloud, Mountain, Natural, River, Sea_Ice, Snow </strong>and <strong>Water.</strong> More details about the competition setup and solution evaluation are on the Kelvins competition platform. The following publication outlines the generation of the dataset: Derksen, D., Meoni, G., Lecuyer, G., Mergy, A., Märtens, M. and Izzo, D. Few-Shot Image Classification Challenge On-Board. NEURIPS2021