Google 致力于为黑人社区推动种族平等。查看具体举措
此页面由 Cloud Translation API 翻译。
Switch to English

从旧的自定义模型API迁移

Firebase/MLModelInterpreter库的0.20.0版引入了一个新的getLatestModelFilePath()方法,该方法获取自定义模型在设备上的位置。您可以使用此方法直接实例化TensorFlow Lite Interpreter对象,您可以使用该对象代替Firebase的ModelInterpreter包装器。

展望未来,这是首选方法。由于TensorFlow Lite解释器版本不再与Firebase库版本配合使用,因此您可以在需要时拥有更大的灵活性来升级到TensorFlow Lite的新版本,或者更轻松地使用自定义TensorFlow Lite构建。

该页面显示了如何从使用ModelInterpreter迁移到TensorFlow Lite Interpreter

1.更新项目依赖项

更新项目的Podfile,以包含Firebase/MLModelInterpreter库(或更高版本)和TensorFlow Lite库的0.20.0版本:

之前

迅速

pod 'Firebase/MLModelInterpreter', '0.19.0'

物镜

pod 'Firebase/MLModelInterpreter', '0.19.0'

迅速

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

物镜

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

2.创建一个TensorFlow Lite解释器,而不是Firebase ModelInterpreter

无需创建Firebase ModelInterpreter ,而是使用getLatestModelFilePath()获取设备上模型的位置,然后使用它来创建TensorFlow Lite Interpreter

之前

迅速

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

目标C

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

迅速

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?
    }
}

物镜

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.更新输入和输出准备代码

使用ModelInterpreter ,您可以通过在运行模型时将ModelInputOutputOptions对象传递给解释器来指定模型的输入和输出形状。

对于TensorFlow Lite解释器,您可以调用allocateTensors()为模型的输入和输出分配空间,然后将输入数据复制到输入张量。

例如,如果模型的输入形状为[1 224 224 3] float值,而输出形状为[1 1000] float值,请进行以下更改:

之前

迅速

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
    // ...
}

目标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
  // ...
}];

迅速

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)
}

目标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.更新输出处理代码

最后,不要使用ModelOutputs对象的output()方法获取模型的输出,而是从解释器获取输出张量并将其数据转换为适合您的使用案例的任何结构。

例如,如果您要进行分类,则可以进行如下更改:

之前

迅速

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)")
    }
}

目标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);
}

迅速

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)
}

目标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]);
}