{
  "schema_version": "rangoon.article.v1",
  "id": "https://rangoon.ai/insights/archive-meta-sam-3-sam-3d-november-2025/",
  "url": "https://rangoon.ai/insights/archive-meta-sam-3-sam-3d-november-2025/",
  "markdown_url": "https://rangoon.ai/insights/archive-meta-sam-3-sam-3d-november-2025/index.md",
  "json_url": "https://rangoon.ai/insights/archive-meta-sam-3-sam-3d-november-2025/index.json",
  "title": "Meta paired concept segmentation with 3D reconstruction in its November 2025 SAM release",
  "summary": "Meta’s November 19, 2025 SAM release paired text- and example-prompted concept segmentation with new 3D object and human-body reconstruction research artifacts.",
  "language": "en",
  "author": {
    "name": "Rangoon Editorial",
    "url": "https://rangoon.ai/insights/editorial/"
  },
  "publisher": {
    "name": "Rangoon",
    "maintainer": "Hypler",
    "url": "https://rangoon.ai/"
  },
  "format": "News analysis",
  "topic": {
    "slug": "models",
    "label": "Models",
    "url": "https://rangoon.ai/insights/topics/models/"
  },
  "tags": [
    {
      "slug": "meta-ai",
      "label": "Meta AI",
      "url": "https://rangoon.ai/insights/tags/meta-ai/"
    },
    {
      "slug": "segment-anything",
      "label": "Segment Anything",
      "url": "https://rangoon.ai/insights/tags/segment-anything/"
    },
    {
      "slug": "computer-vision",
      "label": "computer vision",
      "url": "https://rangoon.ai/insights/tags/computer-vision/"
    },
    {
      "slug": "3d-reconstruction",
      "label": "3D reconstruction",
      "url": "https://rangoon.ai/insights/tags/3d-reconstruction/"
    }
  ],
  "dates": {
    "published": "2026-10-24",
    "updated": "2026-10-24",
    "event": "2025-11-19"
  },
  "event_source_id": 1,
  "retrospective": true,
  "reading_minutes": 3,
  "takeaways": [
    "The November event joined two different research tracks: promptable concept segmentation and image-to-3D reconstruction.",
    "SAM 3’s reported concept-segmentation gains and SAM 3D’s reconstruction claims are release-era research results under their published evaluations.",
    "The records describe models, code, checkpoints and benchmarks; they do not establish safe performance for a specific production, robotics or medical use."
  ],
  "source_note": "Historical reporting: Meta’s November 19, 2025 SAM 3D launch post said it launched alongside SAM 3. The accompanying Meta research records describe separate concept-segmentation and 3D-reconstruction artifacts. The event is a research and artifact release, not a claim that every image yields a correct reconstruction, that a system understands physical interactions, or that a robotics or medical workflow is certified.",
  "sources": [
    {
      "id": 1,
      "title": "Meta AI: Introducing SAM 3D: Powerful 3D Reconstruction for Physical World Images",
      "url": "https://ai.meta.com/blog/sam-3d/",
      "published": "2025-11-19"
    },
    {
      "id": 2,
      "title": "Meta AI Research: SAM 3: Segment Anything with Concepts",
      "url": "https://ai.meta.com/research/publications/sam-3-segment-anything-with-concepts/",
      "published": "2025-11-19"
    },
    {
      "id": 3,
      "title": "Meta AI Research: SAM 3D: 3Dfy Anything in Images",
      "url": "https://ai.meta.com/research/publications/sam-3d-3dfy-anything-in-images/",
      "published": "2025-11-19"
    }
  ],
  "image": {
    "url": "https://rangoon.ai/assets/editorial/archive-meta-sam-3-sam-3d-november-2025-diagram-v1.webp",
    "thumbnail_url": "https://rangoon.ai/assets/editorial/archive-meta-sam-3-sam-3d-november-2025-diagram-v1-thumb.webp",
    "alt": "Original research explanatory diagram, Separate vision research paths. Labels: Concept prompt, SAM 3, Masks, Image, SAM 3D, Mesh. Source boundary: dated primary release reporting; original editorial analysis.",
    "type": "original-diagram",
    "caption": "Separate vision research paths. Original Rangoon explanatory diagram: Source boundary: dated primary release reporting; original editorial analysis. No vendor interface, photograph, measurement result, or endorsement implied.",
    "credit": "Rangoon / Hypler",
    "source_url": "https://rangoon.ai/brand/",
    "rights_url": "https://rangoon.ai/terms/#media-use",
    "rights_note": "Hypler artwork permissions do not grant rights to third-party marks or imply endorsement."
  },
  "sections": [
    {
      "id": "release-event",
      "url": "https://rangoon.ai/insights/archive-meta-sam-3-sam-3d-november-2025/#release-event",
      "heading": "The release put SAM 3 and SAM 3D on one date while keeping their tasks distinct",
      "paragraphs": [
        "On November 19, 2025, Meta introduced SAM 3D and said the release launched alongside SAM 3. The launch post described SAM 3D Objects for object and scene reconstruction and SAM 3D Body for human body and shape estimation, with checkpoints and inference code. The same post pointed to a playground for trying the new models. That makes the date a public artifact-release event. It does not make concept segmentation, object reconstruction and body estimation one interchangeable capability, because each task has different inputs, outputs and error modes.",
        "Meta’s separate SAM 3 research record described a model that detects, segments and tracks matching objects in images and video from short noun phrases, image examples or both. That record calls the task Promptable Concept Segmentation and says the system returns masks and identities for matching instances. A text prompt can therefore define what a user is seeking without proving that a visual category is complete, correctly named or appropriate for a downstream decision. The release supports a narrower account: Meta published related vision artifacts on one day while documenting separate technical scopes."
      ],
      "bullets": [],
      "source_ids": [
        1,
        2
      ]
    },
    {
      "id": "concept-segmentation",
      "url": "https://rangoon.ai/insights/archive-meta-sam-3-sam-3d-november-2025/#concept-segmentation",
      "heading": "SAM 3’s reported concept work depended on a new data engine and benchmark",
      "paragraphs": [
        "Meta reported a twofold gain over existing systems in image and video promptable concept segmentation and released the SA-Co benchmark with SAM 3. That is a vendor-published research comparison, not a blanket performance guarantee. A benchmark can make a defined task measurable while leaving questions about a new camera, rare object, lighting condition, vocabulary, threshold or human consequence unresolved. The November event established that the model and benchmark were opened to researchers; it did not independently validate a customer’s detection pipeline or a safety-critical use."
      ],
      "bullets": [],
      "source_ids": [
        2
      ]
    },
    {
      "id": "reconstruction",
      "url": "https://rangoon.ai/insights/archive-meta-sam-3-sam-3d-november-2025/#reconstruction",
      "heading": "SAM 3D separated object reconstruction from human-body estimation",
      "paragraphs": [
        "The SAM 3D announcement describes two models rather than one general physical-world answer. SAM 3D Objects was presented for reconstructing objects and scenes from images, while SAM 3D Body was presented for single-image body and shape estimation. The body model accepts guidance such as segmentation masks and two-dimensional keypoints and uses Meta’s Momentum Human Rig representation. Those are design and release facts, not evidence that a generated mesh carries a person’s identity, medical status or motion intent.",
        "The launch post also records explicit limits. It says moderate output resolution can reduce detail for complex objects, that whole-person reconstruction can distort or lose detail, and that the object system predicts one object at a time rather than reasoning about physical contact or interpenetration. For SAM 3D Body, the post says it does not account for multi-person or human-object interactions and identifies hand-pose work as an area for improvement. Those limits are central to reading the release as research infrastructure rather than as a completed simulation or robotics-control system."
      ],
      "bullets": [],
      "source_ids": [
        1
      ]
    },
    {
      "id": "historical-boundary",
      "url": "https://rangoon.ai/insights/archive-meta-sam-3-sam-3d-november-2025/#historical-boundary",
      "heading": "A release of research artifacts did not certify downstream visual decisions",
      "paragraphs": [
        "Meta made model checkpoints, inference code and evaluation material part of the November release record. Such artifacts can let researchers and developers test a method, compare it with a defined benchmark and inspect where it fails. They do not remove the need to check dataset rights, test on the intended population, set review thresholds or determine whether a visual output may influence a consequential choice. The research claims remain conditional on Meta’s methods and stated experiments, even where the post describes potential use in creative media, science or robotics."
      ],
      "bullets": [],
      "source_ids": [
        1,
        2
      ]
    }
  ],
  "project_context": null,
  "related_links": [],
  "permissions_url": "https://rangoon.ai/permissions.txt",
  "content_hash": "sha256:7b4a82e5ddfe12d97fc545c4912a1871846c5e14c4cd3239654b555470406d34"
}