本文运行环境为XCode9 beta2, iOS 11 beta2

1. CoreML

1. CoreML是苹果在WWDC2017 新发布的Framework，方便了Machine Learning在苹果自家平台的接接入与使用，同时苹果提供了Python的coremltools，方便将各大开源模型训练工具的现有模型转化为MLModel。

2. 模型训练

   • Caffe
   • Karas
   • XGBoost
   • scikit-learn

2. vision

vision是一个新的，强大的，易于使用的框架，是苹果于WWDC 2017上针对CoreML使用所提出的新Framework，能快速有效的用于面部检测、面部特征点、文字、矩形、条形码和物体。

3. 集成机器学习

我们将构建一个通过AVCaptureSession捕获到当前图像，并通过MLModel分析，获取到与图像最匹配的物品名字。
界面大概是

那么我们就正式开始

1. 创建Single工程

2. 可以从苹果的“机器学习”页面下载Inception v3。

3. 在Info.plist中添加Privacy - Camera Usage Description

4. 代码编写

1. 首先我们创建一个AVCaptureSession用来获取摄像头的图像

   
       lazy var avSession: AVCaptureSession = AVCaptureSession()
       lazy var preViewLayer: AVCaptureVideoPreviewLayer = {
       return AVCaptureVideoPreviewLayer(session: self.avSession)
       }()
   
       override func viewDidLoad() {
           super.viewDidLoad()
       
           setupAVSession()
       
           preViewLayer.frame = view.bounds
           self.view.layer.insertSublayer(preViewLayer, at: 0)
       
           avSession.startRunning()
       }
   
   
       fileprivate func setupAVSession() {
       
           guard let device = AVCaptureDevice.default(for: .video) else {
               fatalError("this application cannot be run on simulator")
           }
       
           do {
           
               let input = try AVCaptureDeviceInput(device: device)
               avSession.addInput(input)
           
               let output = AVCaptureVideoDataOutput()
               avSession.addOutput(output)
           
               let queue = DispatchQueue(label: "video queue", qos: .userInteractive)
               output.setSampleBufferDelegate(self, queue: queue)
           } catch let error {
           
               print(error)
           }
   }
   

2. 实现AVCaptureVideoDataOutputSampleBufferDelegate代理

   extension ViewController: AVCaptureVideoDataOutputSampleBufferDelegate {
   
       func captureOutput(_ output: AVCaptureOutput, didOutput sampleBuffer: CMSampleBuffer, from connection: AVCaptureConnection) {
   
       //FIXME: 添加CoreML相关逻辑
       }
   }
   

3. 为工程引入MLModel
   直接拖拽，点击Inceptionv3
   

   
   可以看到模型的详细信息

4. 添加模型处理代码

   lazy var inceptionv3ClassificationRequest: VNCoreMLRequest = {
       // Load the ML model through its generated class and create a Vision request for it.
       do {
           let model = try VNCoreMLModel(for: Inceptionv3().model)
           return VNCoreMLRequest(model: model, completionHandler: self.inceptionv3ClassificationHandler)
       } catch {
           fatalError("can't load Vision ML model: \(error)")
       }
   }()
       
   
   extension ViewController {
   
       func inceptionv3ClassificationHandler(request: VNRequest, error: Error?) {
           guard let observations = request.results as? [VNClassificationObservation]
               else { fatalError("unexpected result type from VNCoreMLRequest") }
       
           guard let best = observations.first
            else { fatalError("can't get best result") }
       
           DispatchQueue.main.async {
               print("Classification: \"\(best.identifier)\" Confidence: \(best.confidence)")
               self.classifyLabel.text = best.identifier
           }
       }
   }
   

5. 传入MLModel参数

   func captureOutput(_ output: AVCaptureOutput, didOutput sampleBuffer: CMSampleBuffer, from connection: AVCaptureConnection) {
       guard let pixelBuffer = CMSampleBufferGetImageBuffer(sampleBuffer) else {
           return
       }
       
       let handler = VNImageRequestHandler(cvPixelBuffer: pixelBuffer, options: [:])
       
       do {
           try handler.perform([inceptionv3ClassificationRequest])
       } catch _ {
           
       }
   }
   

4. 效果展示

至此已完成了机器学习的集成，代码已上传到git

参考资料

1. WWDC Session 506
2. WWDC Session 703
3. WWDC Session 710