Training datasets (TDS) are a critical element for most machine learning workflows in Earth Observation (EO).
Unfortunately, the availability of TDS for the EO community is generally limited, available datasets on cover few application areas and they are often limited in geographical/temporal/thematic scope. The absence of a central and long-term repository for TDS and the lack of adopted FAIR data principles further amplified this widely recognised bottleneck, limiting innovation and productivity through ML in EO.
This tender seeks to implement a Training Dataset Platform (TDS-Platform) as a central resource and capability for the EO (and related) communities when working with training (and validation) data in EO. The TDS-Platform will combine a systematic curation facility with open data engineering tools for creating, optimising and analysing TDS. It will link to other EO platform resources to provide advanced EO feature engineering and will allow to train computationally costly ML models based on an optimised cloud based computational architecture. Developing an attractive data offering strategy and establishing an incentive for contributing and making dataset available on the platform is part of the core objectives.
Learn more about this Invitation To Tender on the esa-star Publication page.