Scanning objects using Object Capture

At a glance

Project structure

Source bundle/
└── GuidedCaptureSample/
    └── GuidedCaptureSample/
        ├── AppDataModel.swift  # AppDataModel, ModelState, CaptureMode
        ├── CaptureFolderManager.swift  # CaptureFolderManager, Error
        ├── Views/
        │   ├── ContentView.swift  # ContentView
        │   ├── CaptureOverlayView.swift  # CaptureOverlayView, TutorialView, LocalizedString
        │   ├── ModelView.swift  # ModelView, ARQuickLookController, Coordinator
        │   └── PlayerView.swift  # PlayerView, Coordinator, AVPlayerView
        └── GuidedCaptureSampleApp.swift  # GuidedCaptureSampleApp

Structure observations

• The runtime boundary is WindowGroup; the tree is pruned to lifecycle, state, framework, rendering, and ECS files.
• The source is Swift; role-named types and feature folders separate the major responsibilities.

Overall architecture

flowchart LR
    GuidedCaptureSampleApp_1["GuidedCaptureSampleApp"]
    WindowGroup_2["WindowGroup"]
    ContentView_3["ContentView"]
    AppDataModel_4["AppDataModel"]
    Object_Capture_session_pipeline_5["Object Capture session pipeline"]
    GuidedCaptureSampleApp_1 --> WindowGroup_2
    WindowGroup_2 --> ContentView_3
    ContentView_3 --> AppDataModel_4
    AppDataModel_4 --> Object_Capture_session_pipeline_5

Reference code

GuidedCaptureSample/GuidedCaptureSample/GuidedCaptureSampleApp.swift:11 — executable or app-lifecycle anchor.

struct GuidedCaptureSampleApp: App {
    static let subsystem: String = "com.example.apple-samplecode.guided-capture-sample"

    var body: some Scene {
        WindowGroup {
            ContentView()
                .environment(AppDataModel.instance)
        }
    }
}

The diagram is a responsibility flow, not a claim that every adjacent node directly calls the next. It separates lifecycle, presentation, stable state, framework/session work, and the RealityKit entity graph.

Ownership and state

classDiagram
    AppDataModel --> PhotogrammetrySession : photogrammetrySession
    AppDataModel --> CaptureFolderManager : captureFolderManager
    AppDataModel *-- AppDataModel : instance
    AppDataModel --> CaptureMode : captureMode

Ownership evidence

GuidedCaptureSample/GuidedCaptureSample/AppDataModel.swift:35 — representative stored state or nearest verified lifecycle anchor.

    private(set) var photogrammetrySession: PhotogrammetrySession?

    /// When we start a new capture, the folder will be set here.
    private(set) var captureFolderManager: CaptureFolderManager?

    /// Shows whether the user decided to skip reconstruction.
    private(set) var isSaveDraftEnabled = false

Ownership is intentionally narrow: environment, bindings, observed objects, weak links, and delegate callbacks do not prove lifetime ownership. Initialized stored services and wrapper-managed state do; entities added inside a RealityKit content closure belong to that entity graph.

Class and protocol design

Verified protocol-oriented seams: BottomOverlayButtons → OverlayButtons, TopOverlayButtons → OverlayButtons. Framework conformances remain adapters, not proof of an app-wide protocol architecture.

Access control

No reviewed Swift access example uses fileprivate, public, open; declarations without a modifier are internal.

Reference code

GuidedCaptureSample/GuidedCaptureSample/AppDataModel.swift:35 — representative visibility boundary.

    private(set) var photogrammetrySession: PhotogrammetrySession?

    /// When we start a new capture, the folder will be set here.
    private(set) var captureFolderManager: CaptureFolderManager?

    /// Shows whether the user decided to skip reconstruction.
    private(set) var isSaveDraftEnabled = false

    var messageList = TimedMessageList()

    enum ModelState {
        // ...
    }

Logic ownership and placement

Design patterns

Naming conventions

• Role suffixes are evidence, not decoration: App: GuidedCaptureSampleApp; Model: AppDataModel; Coordinator: Coordinator; Controller: ARQuickLookController; Manager: CaptureFolderManager.
• Protocols: OverlayButtons.
• Commands use verb-led methods: endCapture, removeCaptureFolder, setShowOverlaySheets, saveDraft, attachListeners, detachListeners, handleAppTermination, startNewCapture.
• Files and folders use primary-type or responsibility names such as Views, Models, Rendering, Components, Systems, Packages, or feature names where those boundaries exist.

Architecture takeaways

• Treat GuidedCaptureSampleApp as the owner of executable or scene lifecycle, not automatically as the owner of every entity created later.
• Keep view/controller input near presentation, but move sessions, reconstruction, capture, rendering, shared game state, or documents into stable owners when their lifetime exceeds one callback or render pass.
• Tie asynchronous session/output consumers to an explicit owner so cancellation and teardown follow the same lifecycle.

Source map

• Official Apple documentation