Comparing Core ML and TensorFlow Performance and API Usage on iOS

At WWDC Apple announced Core ML, a framework that allows developers to integrate pre‑trained machine learning models into iOS apps without handling the training process.

Core ML works by converting models (e.g., from Caffe) with coremltools into the .mlmodel format; Xcode then generates Swift/Objective‑C wrapper classes that expose simple APIs for inference.

In contrast, TensorFlow is an open‑source library that provides both model definition and training capabilities using data‑flow graphs, supporting Python and C++.

The author compared the two on the same iOS device using identical datasets. Performance (recognition speed) was similar, but TensorFlow consumed more CPU/GPU power, causing the device to heat up, whereas Core ML was more efficient.

Core ML’s API is straightforward: after adding the .mlmodel to the project, Xcode creates classes such as Resnet50 , Resnet50Input , and Resnet50Output that handle input preprocessing and output parsing. Example code shows how to obtain a prediction with let prediction = try? Resnet50().prediction(image: pixelBuffer) .

@objc class Resnet50:NSObject {
    var model: MLModel
    init(contentsOf url: URL) throws {
        self.model = try MLModel(contentsOf: url)
    }
    convenience override init() {
        let bundle = Bundle(for: Resnet50.self)
        let assetPath = bundle.url(forResource: "Resnet50", withExtension:"mlmodelc")
        try! self.init(contentsOf: assetPath!)
    }
    func prediction(input: Resnet50Input) throws -> Resnet50Output {
        let outFeatures = try model.prediction(from: input)
        let result = Resnet50Output(classLabelProbs: outFeatures.featureValue(for: "classLabelProbs")!.dictionaryValue as! [String : Double], classLabel: outFeatures.featureValue(for: "classLabel")!.stringValue)
        return result
    }
    func prediction(image: CVPixelBuffer) throws -> Resnet50Output {
        let input_ = Resnet50Input(image: image)
        return try self.prediction(input: input_)
    }
}

TensorFlow on iOS requires compiling the C++ library, loading the protobuf graph, and manually handling image tensors and session execution, which is more complex and less convenient.

The article concludes that Core ML lowers the barrier to mobile machine‑learning deployment, while TensorFlow offers more flexibility for model updates and training but at the cost of higher integration effort.