NASA’s Jet Propulsion Laboratory (JPL) is on a mission to find signs of life in our solar system, focusing on the presence of water, a vital element for life. The Ocean World Life Surveyor (OWLS) project at JPL is preparing to send a spacecraft to either Europa, a moon of Jupiter, or Enceladus, a moon of Saturn, where ice and water vapor have been discovered. The spacecraft will be equipped with microscopes to collect video data from water samples, looking for evidence of microbes. However, sending this microscopy data back to Earth is a complex and costly task due to the vast distance. Traditional compression methods are inadequate, and the energy cost of downlinking the data is extremely high. The Machine Learning Instrument Autonomy (MLIA) group at JPL faced the challenge of building a machine learning (ML) model that could identify videos most likely to contain signs of life, capture short clips, and prioritize them for downlinking back to Earth.

Creating high-quality training data for machine learning (ML) use cases can be expensive and time-consuming, especially for specialized fields that require domain experts to review and label the data. This is particularly true in the medical field, where doctors are often required to meticulously label or review training data. This process can be arduous and costly, slowing down the development of potentially life-saving algorithms. The challenge was to find a way to create high-quality training data with less involvement from physicians, thereby reducing costs and speeding up the development process.

Creating high-quality video content, such as movies, video games, or live broadcasts, is typically expensive and requires top-notch equipment and animation talent. Traditional technologies for motion capture data collection require markers, which can be bulky, costly, and struggle to capture detailed data. Move.ai, an emerging AI company, aimed to make processes like motion capture and key point recognition easier, faster, and less expensive for both individuals and studios. However, their markerless motion capture solution and AI depth keying tool presented unique challenges when training their AI models. They needed to train algorithms to track a target across keyframes, identify that person or object, identify the people and objects that come into contact with this target, and extract the spatial data of the area captured within the frame.

Sharper Shape, a company that creates technology for safe, efficient transmission and distribution solutions for utilities, was facing challenges in developing their machine learning (ML) models. The company uses computer vision models in advanced aerial sensor systems to power the automatic collection and analysis of unmanned aerial inspection data. A common use case for their technology is the identification of dangerous setups with electric wiring, such as vegetation growing too close, broken insulators, and more. However, training multiple computer vision models required a vast amount of accurately labeled images. Prior to using Labelbox, the Sharper Shape team relied on heavily manual workflows and experimented with open-source labeling tools that did not provide the required amount of configuration needed for their needs. Additionally, each data scientist had spent up to a third of their time on infrastructure and experiment management.