ML Kit AutoML Vision Edge migration guide

Note: The AutoML Vision Edge image labeling API has been deprecated. The custom model APIs (image labeling and object detection and tracking) support models trained by AutoML. This migration guide explains how to migrate an existing AutoML Vision Edge implementation to use the new custom model API.

You can pass an AutoML-trained image classification model to the custom model APIs. You can continue to either bundle the model inside your app or host it on Firebase Console as a custom model. The AutoML image labeling API has been removed from ML Kit since it’s fully replaced by the Custom Model Image Labeling API.

API	What is changing?
AutoML Vision Edge image labeling API	It’s fully replaced by the Custom Model image labeling API. Existing AutoML Vision Edge image labeling API is removed.

If you are currently an ML Kit user using AutoML Vision Edge API, please follow the migration instructions for Android and iOS.

Frequently Asked Questions

Why this change?

It helps simplify ML Kit APIs, and make it easier to integrate ML Kit into your app. With this change, you can use an AutoML-trained model in the exact same way as a custom model. It also enables you to use AutoML-trained models for Object Detection and Tracking, in addition to Image Labeling which we currently support. Furthermore, the custom model API supports both models with label map embedded in its metadata, and models with separate manifest and label files.

What benefits do I get from migrating to the new SDK?

• New features: Ability to use AutoML-trained models for both Image Labeling and Object Detection & Tracking and ability to use models with label map embedded in its metadata.

Migration Guide for Android

Step 1: Update Gradle Imports

Update the dependencies for the ML Kit Android libraries in your module (app-level) Gradle file (usually app/build.gradle) according to the following table:

Feature	Old Artifacts	New Artifact
Image labeling AutoML without remote model downloading	com.google.mlkit:image-labeling-automl:16.2.1	com.google.mlkit:image-labeling-custom:16.0.0-beta5
Image labeling AutoML with remote model downloading	com.google.mlkit:image-labeling-automl:16.2.1 com.google.mlkit:linkfirebase:16.0.1	com.google.mlkit:image-labeling-custom:16.0.0-beta5 com.google.mlkit:linkfirebase:17.0.0

Step 2: Update class names

If your class appears in this table, make the indicated change:

Old class	New class
com.google.mlkit.vision.label.automl.AutoMLImageLabelerLocalModel	com.google.mlkit.common.model.LocalModel
com.google.mlkit.vision.label.automl.AutoMLImageLabelerRemoteModel	com.google.mlkit.common.model.CustomRemoteModel
com.google.mlkit.vision.label.automl.AutoMLImageLabelerOptions	com.google.mlkit.vision.label.custom.CustomImageLabelerOptions

Note: Previously with AutoMLImageLabelerOptions, if the confidenceThreshold is not set, a default confidence threshold value of 0.5 will be used. Now with CustomImageLabelerOptions, if the confidenceThreshold is not set, it will use the confidence threshold value included in the model metadata, if available. If that does not exist, a value of 0.0 will be used instead.

Step 3: Update method names

There are minimal code changes:

• LocalModel can now be initialized with either a model file path (if the model has metadata containing the label map) or a model manifest file path (if the manifest, model, and labels are in separate files).
• You can host a custom model remotely via Firebase Console and initialize a CustomRemoteModel with a FirebaseModelSource.

Here are some examples of old and new Kotlin methods:

Old

val localModel = AutoMLImageLabelerLocalModel.Builder()
    .setAssetFilePath("automl/manifest.json")
    // or .setAbsoluteFilePath(absolute file path to manifest file)
    .build()

val optionsWithLocalModel = AutoMLImageLabelerOptions.Builder(localModel)
    .setConfidenceThreshold(0.5f)
    .build()

val remoteModel = AutoMLImageLabelerRemoteModel.Builder("automl_remote_model")
    .build()

val optionsWithRemoteModel = AutoMLImageLabelerOptions.Builder(remoteModel)
    .build()

New

val localModel = LocalModel.Builder()
    .setAssetManifestFilePath("automl/manifest.json")
    // or .setAbsoluteManifestFilePath(absolute file path to manifest file)
    .build()

val optionsWithLocalModel = CustomImageLabelerOptions.Builder(localModel)
    .setConfidenceThreshold(0.5f)
    .build()

val firebaseModelSource = FirebaseModelSource.Builder("automl_remote_model")
    .build()
val remoteModel = CustomRemoteModel.Builder(firebaseModelSource).build()
val optionsWithRemoteModel = CustomImageLabelerOptions.Builder(remoteModel)
    .build()

Here are some examples of old and new Java methods:

Old

AutoMLImageLabelerLocalModel localModel =
    new AutoMLImageLabelerLocalModel.Builder()
        .setAssetFilePath("automl/manifest.json")
        // or .setAbsoluteFilePath(absolute file path to manifest file)
        .build();
AutoMLImageLabelerOptions optionsWithLocalModel =
    new AutoMLImageLabelerOptions.Builder(localModel)
        .setConfidenceThreshold(0.5f)
        .build();
AutoMLImageLabelerRemoteModel remoteModel =
    new AutoMLImageLabelerRemoteModel.Builder("automl_remote_model").build();
AutoMLImageLabelerOptions optionsWithRemoteModel =
    new AutoMLImageLabelerOptions.Builder(remoteModel)
        .build();

New

LocalModel localModel =
    new LocalModel.Builder()
        .setAssetManifestFilePath("automl/manifest.json")
        // or .setAbsoluteManifestFilePath(absolute file path to manifest file)
        .build()
CustomImageLabelerOptions optionsWithLocalModel =
    new CustomImageLabelerOptions.Builder(localModel)
        .setConfidenceThreshold(0.5f)
        .build();
FirebaseModelSource firebaseModelSource =
    new FirebaseModelSource.Builder("automl_remote_model").build();
CustomRemoteModel remoteModel =
    new CustomRemoteModel.Builder(firebaseModelSource).build();
CustomImageLabelerOptions optionsWithRemoteModel =
    new CustomImageLabelerOptions.Builder(remoteModel).build();

Migration Guide for iOS

Prerequisites

• Xcode 13.2.1 or greater is required.

Step 1: Update Cocoapods

Update the dependencies for the ML Kit iOS cocoapods in your app’s Podfile:

Feature	Old pod name(s)	New pod name(s)
Image labeling AutoML without remote model downloading	GoogleMLKit/ImageLabelingAutoML	GoogleMLKit/ImageLabelingCustom
Image labeling AutoML with remote model downloading	GoogleMLKit/ImageLabelingAutoML GoogleMLKit/LinkFirebase	GoogleMLKit/ImageLabelingCustom GoogleMLKit/LinkFirebase

Step 2: Update class names

If your class appears in this table, make the indicated change:

Swift

Old class	New class
AutoMLImageLabelerLocalModel	LocalModel
AutoMLImageLabelerRemoteModel	CustomRemoteModel
AutoMLImageLabelerOptions	CustomImageLabelerOptions

Objective-C

Old class	New class
MLKAutoMLImageLabelerLocalModel	MLKLocalModel
MLKAutoMLImageLabelerRemoteModel	MLKCustomRemoteModel
MLKAutoMLImageLabelerOptions	MLKCustomImageLabelerOptions

Objective-C

Note: Previously with AutoMLImageLabelerOptions, if the confidenceThreshold is not set, a default confidence threshold value of 0.5will be used. Now with CustomImageLabelerOptions, if the confidenceThreshold is not set, it will use the confidence threshold value included in the model metadata, if available. If that does not exist, a value of 0.0 will be used instead.

Step 3: Update method names

There are minimal code changes:

• LocalModel can now be initialized with either a model file path (if the model has metadata containing the label map) or a model manifest file path (if the manifest, model, and labels are in separate files).
• You can host a custom model remotely via Firebase Console and initialize a CustomRemoteModel with a FirebaseModelSource.

Here are some examples of old and new Swift methods:

Old

let localModel =
    AutoMLImageLabelerLocalModel(manifestPath: "automl/manifest.json")
let optionsWithLocalModel = AutoMLImageLabelerOptions(localModel: localModel)
let remoteModel = AutoMLImageLabelerRemoteModel(name: "automl_remote_model")
let optionsWithRemoteModel = AutoMLImageLabelerOptions(remoteModel: remoteModel)

New

guard let localModel = LocalModel(manifestPath: "automl/manifest.json") else { return }
let optionsWithLocalModel = CustomImageLabelerOptions(localModel: localModel)
let firebaseModelSource = FirebaseModelSource(name: "automl_remote_model")
let remoteModel = CustomRemoteModel(remoteModelSource: firebaseModelSource)
let optionsWithRemoteModel = CustomImageLabelerOptions(remoteModel: remoteModel)

Here are some examples of old and new Objective-C methods:

Old

MLKAutoMLImageLabelerLocalModel *localModel =
    [[MLKAutoMLImageLabelerLocalModel alloc]
        initWithManifestPath:"automl/manifest.json"];
MLKAutoMLImageLabelerOptions *optionsWithLocalModel =
    [[MLKAutoMLImageLabelerOptions alloc] initWithLocalModel:localModel];
MLKAutoMLImageLabelerRemoteModel *remoteModel =
    [[MLKAutoMLImageLabelerRemoteModel alloc]
        initWithManifestPath:"automl/manifest.json"];
MLKAutoMLImageLabelerOptions *optionsWithRemoteModel =
    [[MLKAutoMLImageLabelerOptions alloc] initWithRemoteModel:remoteModel];

New

MLKLocalModel *localModel =
    [[MLKLocalModel alloc] initWithManifestPath:"automl/manifest.json"];
MLKCustomImageLabelerOptions *optionsWithLocalModel =
    [[MLKCustomImageLabelerOptions alloc] initWithLocalModel:localModel];
MLKFirebaseModelSource *firebaseModelSource =
    [[MLKFirebaseModelSource alloc] initWithName:@"automl_remote_model"];
MLKCustomRemoteModel *remoteModel =
    [[MLKCustomRemoteModel alloc] initWithRemoteModelSource:firebaseModelSource];
MLKCustomImageLabelerOptions *optionsWithRemoteModel =
    [[MLKCustomImageLabelerOptions alloc] initWithRemoteModel:remoteModel];