The OPS-SAT case: test dataset

<p><i>The OPS-SAT case: test dataset</i> is the test split of the official dataset of&nbsp;<a href="https://kelvins.esa.int/opssat/home/"><strong>ESA's Kelvins</strong>&nbsp;<strong>"the OPS-SAT case" challenge</strong></a><strong>, </strong>created in collaboration<strong>&nbsp;with ESA's Φ-lab </strong>and <strong>ESA's&nbsp;OPS-SAT spacecraft operations team</strong>. It consists of<strong> 588 raw patches</strong> generated by cropping original acquisitions taken by the OPS-SAT cube-sat using its on-board camera. The patches have a resolution of&nbsp;<strong>200x200</strong> and are provided in&nbsp;<strong>.png</strong> format. Each patch was assigned through manual labeling to one of the following landcover classes: <strong>Agricultural, Cloud, Mountain, Natural, River, Sea_Ice, Snow, </strong>and <strong>Water.</strong></p><p>The competition's goal 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. To this goal, competitors were free to use any data, including <a href="https://zenodo.org/records/6524750">The OPS-SAT case dataset</a>, which was released before the start of the competition.</p><p>The OPS-SAT case: test dataset was hidden from the competitors during the competition and was used to evaluate and rank their submissions.&nbsp;</p><p>More details about the competition setup and solution evaluation are on the <a href="https://kelvins.esa.int/opssat/home/">Kelvins competition platform</a>.</p>