機器學習 Tensorflow 筆記（四）：iOS 通過 Core ML 使用 Tensorflow 訓練模型

摘要： Tensorflow是Google推出的人工智慧框架，而Core ML是蘋果推出的人工智慧框架，兩者是有很大的區別，其中Tensorflow是包含了訓練模型和評估模型，Core ML只支援在裝置上評估模型，不能訓練模型。 通常而言我們通過Tensorflow訓練模型，匯出我們訓練的模型pb...

Tensorflow是Google推出的人工智慧框架，而Core ML是蘋果推出的人工智慧框架，兩者是有很大的區別，其中Tensorflow是包含了訓練模型和評估模型，Core ML只支援在裝置上評估模型，不能訓練模型。

通常而言我們通過Tensorflow訓練模型，匯出我們訓練的模型pb，然後轉換了Core ML的格式mlmodel，從而實現在iOS實現評估模型。

系列文章

匯出MNIST訓練模型

安裝tfcoreml

安裝tfcoreml很簡單， github 上也有詳細的介紹。我的Tensorflow是通過Virtualenv安裝的，如果安裝tfcoreml，那麼需要先啟用環境，通過pip安裝即可。

$ cd targetDirectory 
$ source ./bin/activate
(targetDirectory) $ pip install -U tfcoreml

把pb轉換成mlmodel

import tfcoreml as tf_converter

tf_converter.convert(tf_model_path='model/mnist2.pb',
mlmodel_path='my_model.mlmodel',
output_feature_names=['Softmax:0'],input_name_shape_dict={"input/x_input:0":[1,784]})

專案新增mlmodel模型檔案

把.mlmodel檔案新增到專案中

需要保證檔案已經新增到Compile Sources

分析原理

我們可以點選.mlmodel檔案的箭頭，就可以看到一個程式碼檔案，實際上mlmodel是保護了三個類， Mnist , MnistInput , MnistOutput ,通過這幾個類就可以實現模型的評估。

評估模型

do{
let image = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,3,18,18,18,126,136,175,26,166,255,247,127,0,0,0,0,0,0,0,0,0,0,0,0,30,36,94,154,170,253,253,253,253,253,225,172,253,242,195,64,0,0,0,0,0,0,0,0,0,0,0,49,238,253,253,253,253,253,253,253,253,251,93,82,82,56,39,0,0,0,0,0,0,0,0,0,0,0,0,18,219,253,253,253,253,253,198,182,247,241,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,80,156,107,253,253,205,11,0,43,154,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,14,1,154,253,90,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,139,253,190,2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,11,190,253,70,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,35,241,225,160,108,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,81,240,253,253,119,25,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,45,186,253,253,150,27,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,16,93,252,253,187,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,249,253,249,64,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,46,130,183,253,253,207,2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,39,148,229,253,253,253,250,182,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,24,114,221,253,253,253,253,201,78,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,23,66,213,253,253,253,253,198,81,2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,18,171,219,253,253,253,253,195,80,9,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,55,172,226,253,253,253,253,244,133,11,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,136,253,253,253,212,135,132,16,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]
// print shape
for i in 0...783{
if(i % 28 == 0){
print("")
}
print("\(String(format: "% 2x", image[i]))",terminator: "")
}
print("")

let array = try MLMultiArray(shape: [784], dataType: MLMultiArrayDataType.float32)
for i in 0...(image.count-1) {
let value = Double(image[i]) / 255.0
array[i]= NSNumber(floatLiteral: value)
}
let mnistInput = MnistInput(input__x_input__0: array)
let result = try mnist.prediction(input: mnistInput)
for i in 0...(result.Softmax__0.count - 1){
let item = result.Softmax__0[i]
print("\(i) possibility : \(String(format:"%.2f",item.floatValue))")
}
}catch{
print(error)
}

結果：

最可能是值是 5

0 possibility : 0.00
1 possibility : 0.00
2 possibility : 0.00
3 possibility : 0.14
4 possibility : 0.00
5 possibility : 0.86
6 possibility : 0.00
7 possibility : 0.00
8 possibility : 0.00
9 possibility : 0.00

評估模型（改進）

上面是直接使用陣列進行評估模型，下面我們直接讀取圖片檔案來實現評估模型。

let imagePath = Bundle.main.path(forResource: "test_image", ofType: "png")
let imageData:UIImage = UIImage(contentsOfFile: imagePath!)!
imageView.image = imageData
let width = imageData.cgImage?.width
let height = imageData.cgImage?.height

let data:UnsafePointer<UInt8> = CFDataGetBytePtr(imageData.cgImage?.dataProvider?.data!)
var image = Array<Int>()
for i in 0...(width! * height! - 1){
let postion = i*4
image.append(Int(data[postion]))
}

let mnist = Mnist()
do{
// print shape
for i in 0...783{
if(i % 28 == 0){
print("")
}
print("\(String(format: "% 2x", image[i]))",terminator: "")
}
print("")

let array = try MLMultiArray(shape: [784], dataType: MLMultiArrayDataType.float32)
for i in 0...(image.count-1) {
let value = Double(image[i]) / 255.0
array[i]= NSNumber(floatLiteral: value)
}
let mnistInput = MnistInput(input__x_input__0: array)
let result = try mnist.prediction(input: mnistInput)
var text = ""
for i in 0...(result.Softmax__0.count - 1){
let item = result.Softmax__0[i]
text += "\(i) possibility : \(String(format:"%.2f",item.floatValue))\n"
}
label.text = text
}catch{
print(error)
}

新增圖片

測試結果

Demo原始碼

https://github.com/taoweiji/TensorflowIOSDemo