ML-ready Dataset for Identification of Frost in Martian HiRISE Images

Posted on Slack by Mark Wronkiewicz

Hi all, I wanted to make you aware of a new, open ML-ready dataset designed for the application of machine learning techniques to the study of the Martian seasonal frost cycle. It was curated to deepen our knowledge about Mars’ seasonal frost cycle and its impact on the planet’s climate and surface evolution. The dataset contains ~30k labeled remote sensing image tiles sourced from the HiRISE camera onboard the Mars Reconnaissance Orbiter.We’re open sourcing this dataset (with extensive documentation) to facilitate:

  • General ML Education around binary image classification, calibration of model prediction probabilities, error tradeoffs, etc.
  • Exploring Science-ML Overlap through concepts like model interpretability and explainability, scientifically relevant performance metrics, bias/subpopulation assessment, etc.
  • Public Competitions to invite members of the early career community to delve into scientific ML challenges and explore alternative career pathways

This dataset was generated through JPL’s Science Understanding through Data Science (SUDS) effort to encourage work at the intersection of data science and physical science through dedicated research and other activities like short courses, seminars, and dataset releases.