Case Study: NASA Jet Propulsion Laboratory maps Martian frost with Labelbox

NASA Jet Propulsion Laboratory builds Martian frost dataset over multiple iterations with Labelbox

The NASA Jet Propulsion Laboratory (JPL) faced the challenge of building a holistic map of frost on Mars using diverse data from multiple orbiters. The subtle nature of frost patterns, often confounded by surface features, made creating a reliable training dataset for their machine learning model a complex task. To solve this, they turned to Labelbox and employed its Annotate product to structure their data and generate a high-quality training signal.

Using Labelbox Annotate, JPL implemented an iterative, consensus-driven approach where multiple domain experts reviewed imagery, providing annotations and confidence scores. This process created a multi-expert, confidence-scored signal that made subtle frost patterns trainable for their model. The result was a robust frost-map dataset built over many iterations. This dataset will soon be published for the broader scientific community to refine and use, and it will power JPL's ongoing studies into Martian frost for the next two to three years.

NASA Jet Propulsion Laboratory