top of page

Markerless Object Tracking SDK for Real-World Objects

ModelSet turns your 3D model (GLB/CAD/scan) into a production-ready “object anchor” in minutes—no markers, no fiducials. Publish once, then deploy to Unity, WebXR, iOS and Android with enterprise-grade privacy options (public cloud, private cloud, or fully on-prem). Create object anchors in WebXR with ModelSet.

Scan-Agnostic Object Capture Scan-agnostic object capture—import GLB/PLY from Polycam, Scaniverse or any scanner; Vision Fus

How It Works
Model-based object tracking in Unity & WebXR

Upload 3D File

GLP or PLY 

Author at true scale (1 unit = 1 m)

Align up = +Y

Pick Tracking Type

360° View

Side View

Upload, Process, Publish

Enable localization and tracking in mobile, tablets, headsets and robots

Why teams choose ModelSet:
Model-based object tracking

  • Sub-5 cm accuracy. Rock-solid anchoring with low drift in busy, changing environments.

  • Scan-agnostic. Works with CAD, GLB/GLTF, or meshes from your reality-capture pipeline.

  • Fast to live. Create a ModelSet, upload, process, and ship—typically in under 10 minutes.

  • Built to scale. Indoors or outdoors, multi-floor, any lighting.

  • Deploy anywhere. Unity, WebXR (browser), iOS, Android, Vision Pro, Quest.

  • Enterprise security. Choose cloud, private cloud, or on-prem/self-hosted.

  • Cross-platform SDKs. Unity • Native iOS & Android • WebXR • ROS bridges.

Scan-Agnostic Object Capture

Scan objects with Polycam, Scaniverse or any scanner that outputs GLB & PLY with textured meshes to MultiSet; Vision Fusion normalizes them into object trackers on one high-fidelity map on a unified coordinate system so teams can use the hardware they already own.

Scan-Agnostic.png

Unity Sample Deep-Dive

Unity object tracking sample: GLB → ModelSet → overlays

ModelSet: Track Any Object In Camera FOV

Build real, shippable AR that recognizes the exact object and locks content precisely where it belongs - no markers required. With sub-5 cm anchors from our markerless object tracking SDK, experiences stay rock-solid across Unity, WebXR object anchors, iOS, and Android, so one project ships everywhere.

SDKs & Open APIs

Unity, native iOS & Android, Meta Quest, WebXR and ROS 2 share one binary. Sample scenes, CLI map-uploader, REST/vps/pose and GraphQL map management integrate with any CI/CD. Zero-install endpoints let WebXR & iOS App Clips launch AR from a QR code.

Unity SDK

Drag-and-drop package, C# helpers and sample scenes - localise a Unity app in under 10 min.

MultiSet Unity SDK

WebXR Kit

Zero-install browser runtime; launch spatial experiences from a QR code in < 1 second.

MultiSet WebXR SDK

Native iOS SDK

Swift Package Manager build; leverages LiDAR when present for sub-cm accuracy.

MultiSet iOS SDK

Quest SDK

Native Oculus package with passthrough samples—perfect for industrial AR on Meta Quest Series.

MultiSet Quest SDK

Native Android SDK

Lightweight AAR with Kotlin examples; 52 ms median pose on Snapdragon 8 Gen 3.

MultiSet Andorid SDK

Custom Viewport

Bring your device or robot—we’ll build the viewport and provide a tailored endpoint.

Contact Us

Technical Specs

  • Input formats: GLB/GLTF (CAD and scans convertible)

  • Tracking modes: 360° View, Side View

  • Precision: Sub-5 cm (typical in recommended conditions)

  • Environments: indoor/outdoor, dynamic lighting, multi-floor, large-scale

  • SDKs: Unity, iOS, Android, WebXR; ROS adapters available

  • Privacy: cloud / private cloud / on-prem; opt-ins

Create Object Anchors in WebXR With ModelSet

  • What does MultiSet AI provide?
    MultiSet AI equips developers with everything they need to build large-scale, location-aware applications—3D mapping tools, a state-of-the-art Visual Positioning System (VPS) SDK, and a unified developer platform.
  • Can my scanned data stay in a private cloud?
    Yes. MultiSet offers on-premises deployments and offline SDKs, ensuring your scan data never leaves your infrastructure.
  • Is MultiSet AI's VPS technology based on platforms like Google Cloud Anchors or Apple World Map?
    No, MultiSet AI's VPS technology is built from the ground up, allowing it to scale to thousands of square feet. It is device-agnostic and platform-independent, providing a versatile and robust solution for various AR applications. Compatible with a wide range of hardware and software, our VPS supports multi-floor environments and integrates with existing scan data, offering a comprehensive solution for complex spatial mapping needs.
  • What are the steps to map a space using MultiSet AI?
    To map a space using MultiSet AI, you can use the MultiSet app on your iPhone Pro or iPad Pro to scan the environment. Alternatively, you can import an existing scan into the platform for further processing and integration. Our platform supports a wide range of devices and scan formats, ensuring flexibility and ease of use for developers. Additionally, our technology supports LiDAR mapping and map stitching, allowing for the creation of detailed 3D spatial maps that can be used for various AR applications.
  • Can I bring in third-party scans?
    Yes, We accept E57 files from providers such as Matterport, Leica, NavVis, XGrid, Faro, Polycam and more, and we also support Matterport MatterPak files.
  • How large can one map be before I need MapSet?
    A single map performs best up to ≈2,500 m². For larger footprints or multi-floor venues - split capture zones into logical sections and join them in a MapSet to preserve centimetre accuracy and fast look-ups.The MultiSet app can capture up to 5,000 sq ft (≈465 m²) in a single session. Larger areas can be broken into multiple sessions and merged later on the developer platform. For imports, a single E57 file can be as large as 50,000 sq ft (≈4,650 m²), and multiple files can be merged.
  • Which scanning methods can I use to create a MultiSet map?
    MultiSet is scan-agnostic. Upload LiDAR point clouds, textured meshes or raw SLAM captures; Vision Fusion normalizes them into a single, compression-optimised map ready for VPS localization.
  • How much overlap should adjacent maps have?
    We recommend 15 – 20 % visual overlap between neighbouring maps. This gives MapSet enough shared features to compute high-precision transforms and guarantee seamless hand-offs.
  • Can I update one area without re-mapping the whole venue?
    Yes. Re-scan just the affected zone, upload the new fragment, and MapSet automatically realigns it while the rest of the venue stays online - no downtime or full rebuild required.
  • How do I geo-reference a map for outdoor or mixed-reality use?
    Record the WGS-84 latitude, longitude, altitude and compass heading of your origin point, then enter those values in the project’s Geo Reference panel. Devices can then feed GPS or UWB HintPosition data for faster, more accurate localization.
  • Does MultiSet detect and correct drift over time?
    Continuous background validation checks feature consistency; if drift exceeds a 1 cm threshold, the system flags the sector for optional re-capture or automatic drift compensation.
  • What is the size of map files and can they be compressed?
    A typical indoor map averages 5–15 MB after Vision Fusion compression. For richer detail, toggle “High-Density.” For faster streaming to mobile devices, choose “Edge-Optimised” for files under 3 MB.
  • Can I export maps to other spatial tools?
    Yes. Maps and MapSets can be exported as OBJ or PLY with embedded transform metadata, letting you reuse geometry in BIM, game engines or digital-twin analytics platforms - no vendor lock-in.
Unity Sample — GLB → ModelSet → Overlays Unity object tracking sample—GLB to ModelSet pipeline locks overlays to the exact o

Contact MultiSet AI

bottom of page