Crimson AI NewsA CrimsonLingua Network service
EN ع
← Back to news
Research paper Meta AI

Meta's SAM 3 and DINOv3 Power Real-Time Scientific Imaging at DOE Labs

AI By Crimson AI Meta AI Blog 25 July 2026 · 08:12 36 views
Share: X Telegram

Meta's open-source AI models, Segment Anything Model 3 (SAM 3) and DINOv3, are being used by the SYNAPS-I project at Lawrence Berkeley National Laboratory to accelerate scientific image analysis from weeks to 15 minutes, enabling real-time interpretation of X-ray and neutron data.

Meta's SAM 3 and DINOv3 Power Real-Time Scientific Imaging at DOE Labs

Key points

Meta's open-source AI models are playing a pivotal role in the first wave of the White House's Genesis Mission, a national initiative to accelerate scientific discovery using AI. Lawrence Berkeley National Laboratory (Berkeley Lab) is leading SYNAPS-I, a multi-lab project that aims to transform data analysis at X-ray and neutron facilities from a months-long bottleneck into a real-time discovery engine.

The challenge is immense: DOE's light and neutron source facilities now produce tens of petabytes of data annually, with upgraded detectors capturing 100,000 images per second—up from one image every six seconds a decade ago. Manual analysis cannot keep pace, especially for segmentation, the task of identifying and delineating structures in images. This process previously consumed weeks of expert effort per dataset.

At the heart of SYNAPS-I's solution are two Meta models: Segment Anything Model 3 (SAM 3) and DINOv3. DINOv3, a self-supervised vision model, learns visual patterns without labeled data and provides global context to identify structures. SAM 3 draws precise pixel-level boundaries around objects. Together, they form a complementary pipeline that fine-tuned on scientific imaging data and deployed across 300 A100 GPUs at national supercomputing facilities like NERSC.

The result is a fully reconstructed, semantically labeled 3D volume delivered to scientists in about 15 minutes—while the experiment is still running. The team demonstrated this on micro-CT scans of grapevine stems to study drought resilience, identifying xylem vessels automatically. What previously required a month of expert annotation per time step now takes 15 minutes.

Meta's open-source approach is critical, as national laboratories must keep prepublication data and AI models on secure government infrastructure. The SYNAPS-I team can download, fine-tune, and deploy SAM and DINO within their own secure environments. With 60 researchers across five national labs, the project aims to create intelligent discovery platforms where AI not only processes data faster but also generates hypotheses and recommends experiments.

DOE Under Secretary Dario Gil highlighted the impact: "By seamlessly combining AI, advanced computing, and experimental systems, SYNAPS-I analyzes data as it's produced and guides experiments in real time... This compresses discovery time from days to moments."

MetricBefore SYNAPS-IAfter SYNAPS-I
Image capture rate1 image every 6 seconds100,000 images per second
Segmentation time per dataset~1 month~15 minutes
Annual data volume (DOE facilities)Tens of petabytes
GPUs used300 A100
Source
Meta AI · Meta AI Blog
Related news
Meta AI
Meta AI 27 Jul 2026

Meta’s AI Models Power Assistive Robotics Platform for Wheelchair Users

Meta’s open-source AI vision models DINO and SAM are being integrated into a new assistive robotics platform called RAMMP, led by...

37
Research paper
Meta AI 25 Jul 2026

Meta AI Launches Muse Spark: A Multimodal Reasoning Model Toward Personal Superintelligence

Meta AI introduces Muse Spark, the first model from Meta Superintelligence Labs, featuring native multimodal reasoning, tool use,...

40
Meta AI
Meta AI 25 Jul 2026

Meta Unveils Advanced AI Scaling Framework and Safety Report for Muse Spark

Meta details its updated Advanced AI Scaling Framework, a new Safety & Preparedness Report for its Muse Spark model, and advances...

42