COSMIC

The Content-based Object Summarization to Monitor Infrequent Change (COSMIC) project aims to enable the future of NASA/JPL orbital autonomy. By training machine learning (ML) algorithms to recognize scientifically interesting landforms that change and evolve over time, a COSMIC orbiter will be able to build an onboard encyclopedia of where and when they were observed. Then, on the next pass, it can compare these notes to new observations to make discoveries like "This impact crater is new," "That looks like a fresh impact just by its pattern," "This gully wasn't actively flowing before," etc. These discoveries will be made on using high-resolution images onboard without first requiring that they be sent to Earth. When interesting landforms or change are detected, scientists on Earth be notified, enabling selective download if desired. This conserves both downlink bandwidth and human time, paving the way for an in-situ planetary change detection system.

Naturally, these ML algorithms require a lot of training data, and that's where your labels come in! We need high-quality boundaries to be drawn around each currently known landform example so the ML algorithms can get very good at recognizing what scientists want to find. There will be many tasks created, each focusing on a single type of interesting landform. We look forward to your contributions!