Data entry ops · Production

NASA JPL uses Labelbox video annotation to train a life-detection ML model for spacecraft missions

The problem

The MLIA team's PhD ML researchers were spending time writing custom Java GUIs and manually sorting through data samples to create labeling solutions for each new project, which was a highly inefficient use of their time and slowed ML development.

First attempt

The in-house approach required PhD-level ML researchers to build Java GUIs from scratch for each project, diverting their expertise away from model development.

Workflow diagram · grounded in source
1
Trigger: label diverse microscopy data
trigger
“the team set about labeling a large, diverse set of data sourced from a variety of field water samples and lab-grown specimens”
2
Upload datasets to Labelbox
integration
“The team quickly uploaded video datasets collected from Digital Holographic Microscopes”
3
Annotate object movements in video
human_review
“labelers were easily able to track various objects' movements over time and determine if it might be a live organism”
4
Scientists review model outputs
human_review
“scientists are integrated into the ML pipeline. The MLIA team coordinates with them to review model outputs and verify whether their model tracks the right objects, folding their expertise into their iterative cycles throughout the process”
5
ML model identifies life-like motion
ai_action
“build an ML model that identifies microscopy videos most likely to contain signs of life-like motion, captures short clips of the potentially living creatures, and prioritizes them to downlink back to Earth along with an explanation of i…”
Reported outcome

With Labelbox, the team achieved 5x the speed of customization and annotation and completed their most manual-intensive setup tasks in a day rather than the budgeted week, while also creating significant improvements in their model.

Reported metrics
Annotation and setup speed5x
Setup time (budgeted vs actual)budgeted a week to get it set up, but we were done in a day
Model quality improvementsignificant improvements in our model
Reported stack
Labelboxdecision treesSVMs
Source
https://labelbox.com/customers/nasa-jpl/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

With Labelbox, the team achieved 5x the speed of customization and annotation and completed their most manual-intensive setup tasks in a day rather than the budgeted week, while also creating significant improvements…

What tools did this team use?

Labelbox, decision trees, SVMs.

What results were reported?

Annotation and setup speed: 5x; Setup time (budgeted vs actual): budgeted a week to get it set up, but we were done in a day; Model quality improvement: significant improvements in our model (source-reported, not independently verified).

What failed first in this deployment?

The in-house approach required PhD-level ML researchers to build Java GUIs from scratch for each project, diverting their expertise away from model development.

How is this data entry ops AI workflow structured?

Trigger: label diverse microscopy data → Upload datasets to Labelbox → Annotate object movements in video → Scientists review model outputs → ML model identifies life-like motion.

Related data entry ops cases