Migrate from the legacy custom model API

Version 0.20.0 of the Firebase/MLModelInterpreter library introduces a new getLatestModelFilePath() method, which gets the location on the device of custom models. You can use this method to directly instantiate a TensorFlow Lite Interpreter object, which you can use instead of Firebase's ModelInterpreter wrapper.

Going forward, this is the preferred approach. Because the TensorFlow Lite interpreter version is no longer coupled with the Firebase library version, you have more flexibility to upgrade to new versions of TensorFlow Lite when you want, or more easily use custom TensorFlow Lite builds.

This page shows how you can migrate from using ModelInterpreter to the TensorFlow Lite Interpreter.

1. Update project dependencies

Update your project's Podfile to include version 0.20.0 of the Firebase/MLModelInterpreter library (or newer) and the TensorFlow Lite library:

Before

Swift

pod 'Firebase/MLModelInterpreter', '0.19.0'

Objective-C

pod 'Firebase/MLModelInterpreter', '0.19.0'

After

Swift

pod 'Firebase/MLModelInterpreter', '~> 0.20.0'
pod 'TensorFlowLiteSwift'

Objective-C

pod 'Firebase/MLModelInterpreter', '~> 0.20.0'
pod 'TensorFlowLiteObjC'

2. Create a TensorFlow Lite interpreter instead of a Firebase ModelInterpreter

Instead of creating a Firebase ModelInterpreter, get the model's location on device with getLatestModelFilePath() and use it to create a TensorFlow Lite Interpreter.

Before

Swift

let remoteModel = CustomRemoteModel(
    name: "your_remote_model"  // The name you assigned in the Firebase console.
)
interpreter = ModelInterpreter.modelInterpreter(remoteModel: remoteModel)

Objective-C

// Initialize using the name you assigned in the Firebase console.
FIRCustomRemoteModel *remoteModel =
        [[FIRCustomRemoteModel alloc] initWithName:@"your_remote_model"];
interpreter = [FIRModelInterpreter modelInterpreterForRemoteModel:remoteModel];

After

Swift

let remoteModel = CustomRemoteModel(
    name: "your_remote_model"  // The name you assigned in the Firebase console.
)
ModelManager.modelManager().getLatestModelFilePath(remoteModel) { (remoteModelPath, error) in
    guard error == nil, let remoteModelPath = remoteModelPath else { return }
    do {
        interpreter = try Interpreter(modelPath: remoteModelPath)
    } catch {
        // Error?
    }
}

Objective-C

FIRCustomRemoteModel *remoteModel =
        [[FIRCustomRemoteModel alloc] initWithName:@"your_remote_model"];
[[FIRModelManager modelManager] getLatestModelFilePath:remoteModel
                                            completion:^(NSString * _Nullable filePath,
                                                         NSError * _Nullable error) {
    if (error != nil || filePath == nil) { return; }

    NSError *tfError = nil;
    interpreter = [[TFLInterpreter alloc] initWithModelPath:filePath error:&tfError];
}];

3. Update input and output preparation code

With ModelInterpreter, you specify the model's input and output shapes by passing a ModelInputOutputOptions object to the interpreter when you run it.

For the TensorFlow Lite interpreter, you instead call allocateTensors() to allocate space for the model's input and output, then copy your input data to the input tensors.

For example, if your model has an input shape of [1 224 224 3] float values and an output shape of [1 1000] float values, make these changes:

Before

Swift

let ioOptions = ModelInputOutputOptions()
do {
    try ioOptions.setInputFormat(
        index: 0,
        type: .float32,
        dimensions: [1, 224, 224, 3]
    )
    try ioOptions.setOutputFormat(
        index: 0,
        type: .float32,
        dimensions: [1, 1000]
    )
} catch let error as NSError {
    print("Failed to set input or output format with error: \(error.localizedDescription)")
}

let inputs = ModelInputs()
do {
    let inputData = Data()
    // Then populate with input data.

    try inputs.addInput(inputData)
} catch let error {
    print("Failed to add input: \(error)")
}

interpreter.run(inputs: inputs, options: ioOptions) { outputs, error in
    guard error == nil, let outputs = outputs else { return }
    // Process outputs
    // ...
}

Objective-C

FIRModelInputOutputOptions *ioOptions = [[FIRModelInputOutputOptions alloc] init];
NSError *error;
[ioOptions setInputFormatForIndex:0
                             type:FIRModelElementTypeFloat32
                       dimensions:@[@1, @224, @224, @3]
                            error:&error];
if (error != nil) { return; }
[ioOptions setOutputFormatForIndex:0
                              type:FIRModelElementTypeFloat32
                        dimensions:@[@1, @1000]
                             error:&error];
if (error != nil) { return; }

FIRModelInputs *inputs = [[FIRModelInputs alloc] init];
NSMutableData *inputData = [[NSMutableData alloc] initWithCapacity:0];
// Then populate with input data.

[inputs addInput:inputData error:&error];
if (error != nil) { return; }

[interpreter runWithInputs:inputs
                   options:ioOptions
                completion:^(FIRModelOutputs * _Nullable outputs,
                             NSError * _Nullable error) {
  if (error != nil || outputs == nil) {
    return;
  }
  // Process outputs
  // ...
}];

After

Swift

do {
    try interpreter.allocateTensors()

    let inputData = Data()
    // Then populate with input data.

    try interpreter.copy(inputData, toInputAt: 0)

    try interpreter.invoke()
} catch let err {
    print(err.localizedDescription)
}

Objective-C

NSError *error = nil;

[interpreter allocateTensorsWithError:&error];
if (error != nil) { return; }

TFLTensor *input = [interpreter inputTensorAtIndex:0 error:&error];
if (error != nil) { return; }

NSMutableData *inputData = [[NSMutableData alloc] initWithCapacity:0];
// Then populate with input data.

[input copyData:inputData error:&error];
if (error != nil) { return; }

[interpreter invokeWithError:&error];
if (error != nil) { return; }

4. Update output handling code

Finally, instead of getting the model's output with the ModelOutputs object's output() method, get the output tensor from the interpreter and convert its data to whatever structure is convenient for your use case.

For example, if you're doing classification, you might make changes like the following:

Before

Swift

let output = try? outputs.output(index: 0) as? [[NSNumber]]
let probabilities = output?[0]

guard let labelPath = Bundle.main.path(
    forResource: "custom_labels",
    ofType: "txt"
) else { return }
let fileContents = try? String(contentsOfFile: labelPath)
guard let labels = fileContents?.components(separatedBy: "\n") else { return }

for i in 0 ..< labels.count {
    if let probability = probabilities?[i] {
        print("\(labels[i]): \(probability)")
    }
}

Objective-C

// Get first and only output of inference with a batch size of 1
NSError *error;
NSArray *probabilites = [outputs outputAtIndex:0 error:&error][0];
if (error != nil) { return; }

NSString *labelPath = [NSBundle.mainBundle pathForResource:@"retrained_labels"
                                                    ofType:@"txt"];
NSString *fileContents = [NSString stringWithContentsOfFile:labelPath
                                                   encoding:NSUTF8StringEncoding
                                                      error:&error];
if (error != nil || fileContents == NULL) { return; }
NSArray<NSString *> *labels = [fileContents componentsSeparatedByString:@"\n"];
for (int i = 0; i < labels.count; i++) {
    NSString *label = labels[i];
    NSNumber *probability = probabilites[i];
    NSLog(@"%@: %f", label, probability.floatValue);
}

After

Swift

do {
    // After calling interpreter.invoke():
    let output = try interpreter.output(at: 0)
    let probabilities =
            UnsafeMutableBufferPointer<Float32>.allocate(capacity: 1000)
    output.data.copyBytes(to: probabilities)

    guard let labelPath = Bundle.main.path(
        forResource: "custom_labels",
        ofType: "txt"
    ) else { return }
    let fileContents = try? String(contentsOfFile: labelPath)
    guard let labels = fileContents?.components(separatedBy: "\n") else { return }

    for i in labels.indices {
        print("\(labels[i]): \(probabilities[i])")
    }
} catch let err {
    print(err.localizedDescription)
}

Objective-C

NSError *error = nil;

TFLTensor *output = [interpreter outputTensorAtIndex:0 error:&error];
if (error != nil) { return; }

NSData *outputData = [output dataWithError:&error];
if (error != nil) { return; }

Float32 probabilities[outputData.length / 4];
[outputData getBytes:&probabilities length:outputData.length];

NSString *labelPath = [NSBundle.mainBundle pathForResource:@"custom_labels"
                                                    ofType:@"txt"];
NSString *fileContents = [NSString stringWithContentsOfFile:labelPath
                                                   encoding:NSUTF8StringEncoding
                                                      error:&error];
if (error != nil || fileContents == nil) { return; }

NSArray<NSString *> *labels = [fileContents componentsSeparatedByString:@"\n"];
for (int i = 0; i < labels.count; i++) {
    NSLog(@"%@: %f", labels[i], probabilities[i]);
}