Promptable Segmentation Foundation Models (Segment Anything)
Models that cut out any object from a point, box or text prompt, trained on very large mask collections.
Tier Grades
Tier A: Frontier AI labs
feasible
Building a 1-billion-mask data engine over 11M images is frontier-lab scale.
Tier B: Indonesian AI labs
stretch
Repeating SA-1B scale is beyond an Indonesian lab; fine-tuning SAM on local imagery is feasible.
Tier C: Institution units such as ITB
feasible
Domain adaptation of SAM, as MedSAM did, fits a university unit.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
Prompted segmentation from a released SAM speeds up labelling and measurement in proofs of concept.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeTraining compute is not in the sources read.
- DataSA-1B: over 1 billion masks on 11M licensed and privacy-respecting images.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 12 authors on the Segment Anything paper (arXiv 2304.02643).
- Time horizonYear.
- Shelf-life signalsA new SAM generation adds video or concept prompts, Distilled SAM variants reach on-device speed
- Pipeline roleTier A releases SAM models and SA-1B; Tier C adapts them to domains such as medicine; Tier D uses them for labelling and measurement in client pilots.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Linked models
SAM 3visionexemplar
MedSAMvisionexemplar
MobileSAMvisionexemplarSAM 1visionexemplarSAM 2visionexemplarSAM 3.1 (Object Multiplex)visionproduced byMedSAM2visionproduced byEdgeTAMvisionproduced by
EfficientViT-SAMvisionproduced by
SlimSAMvisionproduced by
FastSAMvisionproduced by
SAM-HQvisionproduced by
Track-Anythingvisionproduced bySAM ViT-Bvisionproduced byLinks
R&D TopicPromptable Segmentation Foundation Models (Segment Anything)
Use CasesMedical Image SegmentationHealthcare
Who Runs Itnone recorded
Models 14
MedSAMBo Wang Lab, Toronto, visionexemplarSAM 3Meta, visionexemplarMedSAM2Bo Wang Lab, Toronto, visionproduced byEdgeTAMMeta, visionproduced bySAM 2Meta, visionexemplarSAM 3.1 (Object Multiplex)Meta, visionproduced byEfficientViT-SAMMIT Han Lab, visionproduced byMobileSAMKyung Hee University, visionexemplarSAM 1Meta, visionexemplarSAM ViT-Bvisionproduced bySAM-HQETH Zurich, Hong Kong University of Science and Technology, visionproduced bySlimSAMNational University of Singapore, visionproduced byPapers
- Segment AnythingAlexander Kirillov et al., arXiv preprint, 2023 founding
- SAM 3: Segment Anything with ConceptsNicolas Carion et al., arXiv preprint, 2025 state of the art
- SAM 2: Segment Anything in Images and VideosNikhila Ravi et al., arXiv preprint, 2024 state of the art
- Faster Segment Anything: Towards Lightweight SAM for Mobile ApplicationsChaoning Zhang et al., arXiv preprint, 2023 state of the art
- EfficientSAM: Leveraged Masked Image Pretraining for Efficient Segment AnythingYunyang Xiong et al., arXiv preprint, 2023 state of the art
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2304.02643 https://arxiv.org/abs/2304.02643
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Promptable Segmentation Foundation Models (Segment Anything)", https://aurion.semai.id/topics/promptable-segmentation-foundation-models/, records of October 2026.