How Meta AI models are leading Genesis Mission's first project
Key point
Meta's SAM 3 and DINOv3 have enabled real-time analysis of X-ray data at DOE national laboratories.
Details
U.S. Department of Energy (DOE) research facilities generate tens of petabytes of experimental data annually. Upgraded detectors can capture 100,000 images per second, a level impossible to keep up with using traditional manual analysis.
The White House launched the Genesis Mission in late 2025, and one of its core projects is SYNAPS-I. Led by Berkeley Lab and involving Argonne, Brookhaven, Oak Ridge, and SLAC National Laboratories, this initiative aims to convert X-ray and neutron science data analysis into real-time processing.
At the core of SYNAPS-I are two open-source models released by Meta.
- SAM 3 (Segment Anything Model 3): precisely extracts pixel-level boundaries of individual structures within images
- DINOv3: a self-supervised vision model that learns visual patterns without labels, grasping overall structural context
After fine-tuning both models on DOE beamline data, they were deployed on 300 A100 GPUs at the NERSC supercomputing facility. As a result, fully reconstructed 3D volumes and semantic labels are delivered in real time while scientists stand in front of their experimental equipment. Analysis that previously took weeks is now completed while the experiment is still running.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.