Po zainstalowaniu lub zaktualizowaniu podów projektu otwórz projekt Xcode za pomocą
.xcworkspace ML Kit jest obsługiwany w Xcode w wersji 12.4 lub nowszej.
Teraz możesz dodać etykiety do obrazów.
1. Przygotowywanie obrazu wejściowego
Utwórz obiekt VisionImage za pomocą UIImage lub
CMSampleBuffer.
Jeśli używasz UIImage, wykonaj te czynności:
Utwórz obiekt VisionImage za pomocą UIImage. Pamiętaj, by określić prawidłowy .orientation.
Swift
let image = VisionImage(image: UIImage)
visionImage.orientation = image.imageOrientation
Aby oznaczyć etykietami obiekty na obrazie, przekaż obiekt VisionImage do funkcji
Metoda processImage() użytkownika ImageLabeler.
Najpierw pobierz instancję ImageLabeler.
Swift
letlabeler=ImageLabeler.imageLabeler()// Or, to set the minimum confidence required:// let options = ImageLabelerOptions()// options.confidenceThreshold = 0.7// let labeler = ImageLabeler.imageLabeler(options: options)
Objective-C
MLKImageLabeler*labeler=[MLKImageLabelerimageLabeler];// Or, to set the minimum confidence required:// MLKImageLabelerOptions *options =// [[MLKImageLabelerOptions alloc] init];// options.confidenceThreshold = 0.7;// MLKImageLabeler *labeler =// [MLKImageLabeler imageLabelerWithOptions:options];
3. Uzyskiwanie informacji o obiektach oznaczonych etykietami
Jeśli oznaczenie obrazów zostanie oznaczone etykietami, moduł obsługi uzupełniania otrzyma tablicę
ImageLabel obiekty. Każdy obiekt ImageLabel reprezentuje coś, co
oznaczone etykietą na zdjęciu. Model podstawowy obsługuje ponad 400 różnych etykiet.
Możesz uzyskać opis tekstowy każdej etykiety i zindeksować wszystkie etykiety obsługiwane przez
model oraz wskaźnik ufności dopasowania. Na przykład:
Wskazówki dotyczące poprawy skuteczności w czasie rzeczywistym
Jeśli chcesz oznaczać obrazy w aplikacji działającej w czasie rzeczywistym, postępuj zgodnie z tymi instrukcjami
wytycznych dotyczących uzyskiwania najlepszej liczby klatek na sekundę:
Do przetwarzania klatek wideo używaj synchronicznego interfejsu API results(in:) narzędzia do oznaczania obrazów. Zadzwoń do nas
tę metodę z AVCaptureVideoDataOutputSampleBufferDelegate.
captureOutput(_, didOutput:from:), aby synchronicznie pobierać wyniki dotyczące danego filmu
ramki. Zachowaj AVCaptureVideoDataOutput: alwaysDiscardsLateVideoFrames jako true, aby ograniczyć wywołania osoby oznaczającej obrazy. Jeśli nowy
klatka wideo będzie dostępna podczas działania narzędzia do oznaczania obrazów, a zostanie ona usunięta.
Jeśli używasz danych wyjściowych twórcy etykiet do nakładania grafiki na
obrazu wejściowego, najpierw pobierz wynik z ML Kit, a następnie wyrenderuj obraz
i nakładanie nakładek w jednym kroku. W ten sposób renderowanie na powierzchni
tylko raz na każdą przetworzoną ramkę wejściową. Zobacz updatePreviewOverlayViewWithLastFrame.
znajdziesz na przykład w krótkim wprowadzeniu do korzystania z ML Kit.
[null,null,["Ostatnia aktualizacja: 2025-08-29 UTC."],[[["\u003cp\u003eML Kit's image labeling API lets you identify objects in images using a pre-trained model that recognizes over 400 labels.\u003c/p\u003e\n"],["\u003cp\u003eTo use this API, you need to include the \u003ccode\u003eGoogleMLKit/ImageLabeling\u003c/code\u003e pod, create a \u003ccode\u003eVisionImage\u003c/code\u003e object from your image, and then process it with an \u003ccode\u003eImageLabeler\u003c/code\u003e instance.\u003c/p\u003e\n"],["\u003cp\u003eResults are provided as an array of \u003ccode\u003eImageLabel\u003c/code\u003e objects, each containing the label's text, confidence score, and index.\u003c/p\u003e\n"],["\u003cp\u003eFor real-time applications, leverage the synchronous \u003ccode\u003eresults(in:)\u003c/code\u003e API and manage video frame processing efficiently to maintain optimal frame rates.\u003c/p\u003e\n"]]],["ML Kit allows image labeling using a default model with 400+ labels. To begin, include the `GoogleMLKit/ImageLabeling` pod in your Podfile and open the `.xcworkspace` in Xcode. Input images are prepared using `VisionImage` objects, created from either `UIImage` or `CMSampleBuffer`. An `ImageLabeler` instance processes the image, returning `ImageLabel` objects with text, confidence, and index. For real-time performance, use the synchronous `results(in:)` API with video frames, managing the frame rate.\n"],null,["You can use ML Kit to label objects recognized in an image. The default model provided with\nML Kit supports 400+ different labels.\n\n\u003cbr /\u003e\n\n| **Note:** ML Kit iOS APIs only run on 64-bit devices. If you build your app with 32-bit support, check the device's architecture before using this API.\n\nTry it out\n\n- Play around with [the sample app](https://github.com/googlesamples/mlkit/tree/master/ios/quickstarts/vision) to see an example usage of this API.\n\nBefore you begin\n\n1. Include the following ML Kit pods in your Podfile: \n\n ```\n pod 'GoogleMLKit/ImageLabeling', '8.0.0'\n ```\n2. After you install or update your project's Pods, open your Xcode project using its `.xcworkspace`. ML Kit is supported in Xcode version 12.4 or greater.\n\nNow you are ready to label images.\n\n1. Prepare the input image\n\nCreate a [`VisionImage`](/ml-kit/reference/swift/mlkitvision/api/reference/Classes/VisionImage) object using a `UIImage` or a\n`CMSampleBuffer`.\n\nIf you use a `UIImage`, follow these steps:\n\n- Create a [`VisionImage`](/ml-kit/reference/swift/mlkitvision/api/reference/Classes/VisionImage) object with the `UIImage`. Make sure to specify the correct `.orientation`. \n\n Swift \n\n ```text\n let image = VisionImage(image: UIImage)\n visionImage.orientation = image.imageOrientation\n ```\n\n Objective-C \n\n ```objective-c\n MLKVisionImage *visionImage = [[MLKVisionImage alloc] initWithImage:image];\n visionImage.orientation = image.imageOrientation;\n ```\n\nIf you use a `CMSampleBuffer`, follow these steps:\n\n- Specify the orientation of the image data contained in the\n `CMSampleBuffer`.\n\n To get the image orientation: \n\n Swift \n\n ```gdscript\n func imageOrientation(\n deviceOrientation: UIDeviceOrientation,\n cameraPosition: AVCaptureDevice.Position\n ) -\u003e UIImage.Orientation {\n switch deviceOrientation {\n case .portrait:\n return cameraPosition == .front ? .leftMirrored : .right\n case .landscapeLeft:\n return cameraPosition == .front ? .downMirrored : .up\n case .portraitUpsideDown:\n return cameraPosition == .front ? .rightMirrored : .left\n case .landscapeRight:\n return cameraPosition == .front ? .upMirrored : .down\n case .faceDown, .faceUp, .unknown:\n return .up\n }\n }\n \n ```\n\n Objective-C \n\n ```css+lasso\n - (UIImageOrientation)\n imageOrientationFromDeviceOrientation:(UIDeviceOrientation)deviceOrientation\n cameraPosition:(AVCaptureDevicePosition)cameraPosition {\n switch (deviceOrientation) {\n case UIDeviceOrientationPortrait:\n return cameraPosition == AVCaptureDevicePositionFront ? UIImageOrientationLeftMirrored\n : UIImageOrientationRight;\n\n case UIDeviceOrientationLandscapeLeft:\n return cameraPosition == AVCaptureDevicePositionFront ? UIImageOrientationDownMirrored\n : UIImageOrientationUp;\n case UIDeviceOrientationPortraitUpsideDown:\n return cameraPosition == AVCaptureDevicePositionFront ? UIImageOrientationRightMirrored\n : UIImageOrientationLeft;\n case UIDeviceOrientationLandscapeRight:\n return cameraPosition == AVCaptureDevicePositionFront ? UIImageOrientationUpMirrored\n : UIImageOrientationDown;\n case UIDeviceOrientationUnknown:\n case UIDeviceOrientationFaceUp:\n case UIDeviceOrientationFaceDown:\n return UIImageOrientationUp;\n }\n }\n \n ```\n- Create a [`VisionImage`](/ml-kit/reference/swift/mlkitvision/api/reference/Classes/VisionImage) object using the `CMSampleBuffer` object and orientation: \n\n Swift \n\n ```povray\n let image = VisionImage(buffer: sampleBuffer)\n image.orientation = imageOrientation(\n deviceOrientation: UIDevice.current.orientation,\n cameraPosition: cameraPosition)\n ```\n\n Objective-C \n\n ```objective-c\n MLKVisionImage *image = [[MLKVisionImage alloc] initWithBuffer:sampleBuffer];\n image.orientation =\n [self imageOrientationFromDeviceOrientation:UIDevice.currentDevice.orientation\n cameraPosition:cameraPosition];\n ```\n\n2. Configure and run the image labeler To label objects in an image, pass the `VisionImage` object to the `ImageLabeler`'s `processImage()` method.\n\n\u003cbr /\u003e\n\n1. First, get an instance of `ImageLabeler`.\n\nSwift \n\n```swift\nlet labeler = ImageLabeler.imageLabeler()\n\n// Or, to set the minimum confidence required:\n// let options = ImageLabelerOptions()\n// options.confidenceThreshold = 0.7\n// let labeler = ImageLabeler.imageLabeler(options: options)\n```\n\nObjective-C \n\n```objective-c\nMLKImageLabeler *labeler = [MLKImageLabeler imageLabeler];\n\n// Or, to set the minimum confidence required:\n// MLKImageLabelerOptions *options =\n// [[MLKImageLabelerOptions alloc] init];\n// options.confidenceThreshold = 0.7;\n// MLKImageLabeler *labeler =\n// [MLKImageLabeler imageLabelerWithOptions:options];\n```\n\n1. Then, pass the image to the `processImage()` method:\n\nSwift \n\n```swift\nlabeler.process(image) { labels, error in\n guard error == nil, let labels = labels else { return }\n\n // Task succeeded.\n // ...\n}\n```\n\nObjective-C \n\n```objective-c\n[labeler processImage:image\ncompletion:^(NSArray *_Nullable labels,\n NSError *_Nullable error) {\n if (error != nil) { return; }\n\n // Task succeeded.\n // ...\n}];\n```\n\n3. Get information about labeled objects\n\nIf image labeling succeeds, the completion handler receives an array of\n`ImageLabel` objects. Each `ImageLabel` object represents something that was\nlabeled in the image. The base model supports [400+ different labels](/ml-kit/vision/image-labeling/label-map).\nYou can get each label's text description, index among all labels supported by\nthe model, and the confidence score of the match. For example: \n\nSwift \n\n```swift\nfor label in labels {\n let labelText = label.text\n let confidence = label.confidence\n let index = label.index\n}\n```\n\nObjective-C \n\n```objective-c\nfor (MLKImageLabel *label in labels) {\n NSString *labelText = label.text;\n float confidence = label.confidence;\n NSInteger index = label.index;\n}\n```\n\nTips to improve real-time performance\n\nIf you want to label images in a real-time application, follow these\nguidelines to achieve the best framerates:\n\n- For processing video frames, use the `results(in:)` synchronous API of the image labeler. Call this method from the [`AVCaptureVideoDataOutputSampleBufferDelegate`](https://developer.apple.com/documentation/avfoundation/avcapturevideodataoutputsamplebufferdelegate)'s [`captureOutput(_, didOutput:from:)`](https://developer.apple.com/documentation/avfoundation/avcapturevideodataoutputsamplebufferdelegate/1385775-captureoutput) function to synchronously get results from the given video frame. Keep [`AVCaptureVideoDataOutput`](https://developer.apple.com/documentation/avfoundation/avcapturevideodataoutput)'s [`alwaysDiscardsLateVideoFrames`](https://developer.apple.com/documentation/avfoundation/avcapturevideodataoutput/1385780-alwaysdiscardslatevideoframes) as `true` to throttle calls to the image labeler. If a new video frame becomes available while the image labeler is running, it will be dropped.\n- If you use the output of the image labeler to overlay graphics on the input image, first get the result from ML Kit, then render the image and overlay in a single step. By doing so, you render to the display surface only once for each processed input frame. See the [updatePreviewOverlayViewWithLastFrame](https://github.com/googlesamples/mlkit/blob/master/ios/quickstarts/vision/VisionExample/CameraViewController.swift) in the ML Kit quickstart sample for an example."]]
