iOS-CoreML-初探

CoreML 是 Apple 在 WWDC 2017 推出的机器学习框架。但是其到底有什么功能呢?本文对其进行初探。

模型

CoreML 中, Apple 定义了一套自己的模型格式,后缀名为: mimodel,通过 CoreML 框架,以及模型库,可以在 App 层面进行机器学习的功能研发。

CoreML

官网已经提供四个模型库供下载。

iOS-CoreML-Model

Demo

官网提供了一个 Demo,要求 XCode 9 + iOS 11 的环境。

下载下来 Run 了一下,不得不说,Apple 对开发者还是非常友好的,直接将模型文件拖到项目中,Xcode 会自动生成接口文件:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
import CoreML

class MarsHabitatPricerInput : MLFeatureProvider {
var solarPanels: Double
var greenhouses: Double
var size: Double

var featureNames: Set<String> {
get {
return ["solarPanels", "greenhouses", "size"]
}
}

func featureValue(for featureName: String) -> MLFeatureValue? {
if (featureName == "solarPanels") {
return MLFeatureValue(double: solarPanels)
}
if (featureName == "greenhouses") {
return MLFeatureValue(double: greenhouses)
}
if (featureName == "size") {
return MLFeatureValue(double: size)
}
return nil
}

init(solarPanels: Double, greenhouses: Double, size: Double) {
self.solarPanels = solarPanels
self.greenhouses = greenhouses
self.size = size
}
}

class MarsHabitatPricerOutput : MLFeatureProvider {
let price: Double

var featureNames: Set<String> {
get {
return ["price"]
}
}

func featureValue(for featureName: String) -> MLFeatureValue? {
if (featureName == "price") {
return MLFeatureValue(double: price)
}
return nil
}

init(price: Double) {
self.price = price
}
}

@objc class MarsHabitatPricer:NSObject {
var model: MLModel
init(contentsOf url: URL) throws {
self.model = try MLModel(contentsOf: url)
}
convenience override init() {
let bundle = Bundle(for: MarsHabitatPricer.self)
let assetPath = bundle.url(forResource: "MarsHabitatPricer", withExtension:"mlmodelc")
try! self.init(contentsOf: assetPath!)
}
func prediction(input: MarsHabitatPricerInput) throws -> MarsHabitatPricerOutput {
let outFeatures = try model.prediction(from: input)
let result = MarsHabitatPricerOutput(price: outFeatures.featureValue(for: "price")!.doubleValue)
return result
}
func prediction(solarPanels: Double, greenhouses: Double, size: Double) throws -> MarsHabitatPricerOutput {
let input_ = MarsHabitatPricerInput(solarPanels: solarPanels, greenhouses: greenhouses, size: size)
return try self.prediction(input: input_)
}
}

可以看到,主要是定义了输入,输出以及预测的格式,调用的时候,也非常简单,传参即可。

但是这些接口文件并没有在 XCode 左边的文件树中出现。

查了一下,是生成在 DerivedData 目录下,估计是想开发者使用起来更简洁。

运行一下,可以看到,主要功能是对价格进行预测。

iOS-CoreML-Demo

貌似稍微有点不够高大上…

Resnet50

官网提供的四个模型库,我们还没用呢,当然要看下能用来干啥,看了一下,貌似主要是物体识别,OK,代码走起。

先下载模型库 Resnet50, 然后创建一个新的 Swift 项目,将其拖进去:

iOS-CoreML-Model-Resnet50

从描述里面可以看出来,其实一个神经网络的分类器,输入是一张像素为 (224 * 224) 的图片,输出为分类结果。

自动生成的接口文件:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
import CoreML

class Resnet50Input : MLFeatureProvider {
var image: CVPixelBuffer

var featureNames: Set<String> {
get {
return ["image"]
}
}

func featureValue(for featureName: String) -> MLFeatureValue? {
if (featureName == "image") {
return MLFeatureValue(pixelBuffer: image)
}
return nil
}

init(image: CVPixelBuffer) {
self.image = image
}
}

class Resnet50Output : MLFeatureProvider {
let classLabelProbs: [String : Double]
let classLabel: String

var featureNames: Set<String> {
get {
return ["classLabelProbs", "classLabel"]
}
}

func featureValue(for featureName: String) -> MLFeatureValue? {
if (featureName == "classLabelProbs") {
return try! MLFeatureValue(dictionary: classLabelProbs as [NSObject : NSNumber])
}
if (featureName == "classLabel") {
return MLFeatureValue(string: classLabel)
}
return nil
}

init(classLabelProbs: [String : Double], classLabel: String) {
self.classLabelProbs = classLabelProbs
self.classLabel = classLabel
}
}

@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_)
}
}

OK,要照片,而且是 CVPixelBuffer 类型的。

但是每次从相册选太烦了,所以我们直接摄像头走起。将 AVCam 的主要功能类复制到项目中。

iOS-CoreML-AVCam

然后,禁用 CameraViewController 中一些不必要的按钮:

1
2
3
4
self.recordButton.isHidden = true
self.captureModeControl.isHidden = true
self.livePhotoModeButton.isHidden = true
self.depthDataDeliveryButton.isHidden = true

由于,AVCapturePhotoCaptureDelegate 拍照完成的回调为:

1
func photoOutput(_ output: AVCapturePhotoOutput, didFinishProcessingPhoto photo: AVCapturePhoto, error: Error?)

看了下 AVCaputrePhoto 的定义,里面刚好有 CVPixelBuffer 格式的属性:

iOS-CoreML-AVCam-PreviewPixelBuffer

直接传进去试试:

1
2
3
4
5
6
// Predicte
if let pixelBuffer = photo.previewPixelBuffer {
guard let Resnet50CategoryOutput = try? model.prediction(image:pixelBuffer) else {
fatalError("Unexpected runtime error.")
}
}

一切看起来很完美,编译通过,运行起来,点一下拍照按钮,额,Crash了,异常:

1
[core] Error Domain=com.apple.CoreML Code=1 "Input image feature image does not match model description" UserInfo={NSLocalizedDescription=Input image feature image does not match model description, NSUnderlyingError=0x1c0643420 {Error Domain=com.apple.CoreML Code=1 "Image is not valid width 224, instead is 852" UserInfo={NSLocalizedDescription=Image is not valid width 224, instead is 852}}}

哦,忘记改大小了,找到 photoSetting,加上宽高:

1
2
3
4
5
if !photoSettings.availablePreviewPhotoPixelFormatTypes.isEmpty {
photoSettings.previewPhotoFormat = [kCVPixelBufferPixelFormatTypeKey as String: photoSettings.availablePreviewPhotoPixelFormatTypes.first!,
kCVPixelBufferWidthKey as String : NSNumber(value:224),
kCVPixelBufferHeightKey as String : NSNumber(value:224)]
}

重新 Run,WTF,Man,居然又报同样的错,好吧,Google 一下,貌似宽高的属性,在 Swift 里面不生效,额。。

没办法,那我们只能将 CVPixelBuffer 先转换成 UIImage,然后改下大小,再转回 CVPixelBuffer,试试:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
photoData = photo.fileDataRepresentation()

// Change Data to Image
guard let photoData = photoData else {
return
}
let image = UIImage(data: photoData)

// Resize
let newWidth:CGFloat = 224.0
let newHeight:CGFloat = 224.0
UIGraphicsBeginImageContext(CGSize(width:newWidth, height:newHeight))
image?.draw(in:CGRect(x:0, y:0, width:newWidth, height:newHeight))
let newImage = UIGraphicsGetImageFromCurrentImageContext()
UIGraphicsEndImageContext()

guard let finalImage = newImage else {
return
}

// Predicte
guard let Resnet50CategoryOutput = try? model.prediction(image:pixelBufferFromImage(image: finalImage)) else {
fatalError("Unexpected runtime error.")
}

重新 Run,OK,一切很完美。

最后,为了用户体验,加上摄像头流的暂停和重启,免得在识别的时候,摄像头还一直在动,另外,识别结果通过提醒框弹出来,具体参考文末的源码。

开始玩啦,找支油笔试一下:

iOS-CoreML-Pen

识别成,橡皮擦,好吧,其实是有点像。

再拿小绿植试试:

iOS-CoreML-FlowerPot

花瓶,Are you kidding me ??

其实,效果还是蛮不错的。

刚好下午要去上海 CES Asia,一路拍过去玩,想想都有点小激动。

最后,源码奉上,想玩的同学直接下载编译就行了,别忘了 Star~