【CNTK】CNTK学习笔记之图像识别-树叶识别Train and Test

用手机1:1拍的数据集，之后用我前边一篇博客OpenCV修改图片尺寸把数据集转换成28*28大小。再用之前一篇博客OpenCV制作CNTK数据集，提取灰度图像素值写入文件。并从创建好的txt里拿出一些当做测试集。

大概思路就是这样。图片拍的不多，最后训练集179张，测试集14张。最后有实验结果的截图。
数据集截图：

CNTK配置文件：

rootDir = ".."

configDir = "$rootDir$/Config"
dataDir   = "$rootDir$/Data"
outputDir = "$rootDir$/Output"
modelDir  = "$outputDir$/Models"

deviceId = -1

command = train:test

precision = "float"
modelPath = "$modelDir$/02_Convolution"

traceLevel = 1
numMBsToShowResult = 100

#######################################
#  TRAINING CONFIG                    #
#######################################

train = [
    action = "train"

    NDLNetworkBuilder = [
        imageLayout = "cudnn"
        initOnCPUOnly = true
        ndlMacros = "$configDir$/Macros.ndl"
        networkDescription = "$ConfigDir$/leaf.ndl"
    ]

    SGD = [
        epochSize = 60000
        minibatchSize = 32
        #learningRatesPerSample = 0.003125    # TODO
        #momentumAsTimeConstant = 0
        learningRatesPerMB = 0.1*5:0.3
        momentumPerMB = 0*4:0.7
        maxEpochs = 15
    ]

    reader = [
        readerType = "CNTKTextFormatReader"
        file = "$DataDir$/trainimageleaf.txt"
        input = [
            features = [
                dim = 784
                format = "dense"
            ]
            labels = [
                dim = 4
                format = "dense"
            ]
        ]
    ]    
]

#######################################
#  TEST CONFIG                        #
#######################################

test = [
    action = test
    minibatchSize = 15

    reader = [
        readerType = "CNTKTextFormatReader"
        file = "$DataDir$/testimageleaf.txt"
        input = [
            features = [
            dim = 784
            format = "dense"
            ]
            labels = [
                dim = 4
                format = "dense"
            ]
        ]
    ]
]

NDL网络配置（LeNet-5）：

load = ndlMnistMacros

run = DNN

ndlMnistMacros = [
    imageW = 28
    imageH = 28
    labelDim = 4

    features = ImageInput(imageW, imageH, 1, imageLayout=$imageLayout$)
    featScale = Constant(0.00390625)
    featScaled = Scale(featScale, features)
    labels = InputValue(labelDim)
]

DNN=[
    # conv1
    kW1 = 5
    kH1 = 5
    cMap1 = 16
    hStride1 = 1
    vStride1 = 1
    # weight[cMap1, kW1 * kH1 * inputChannels]
    # Conv2DReLULayer is defined in Macros.ndl
    conv1 = Conv2DReLULayer(featScaled, cMap1, 25, kW1, kH1, hStride1, vStride1, 10, 1)

    # pool1
    pool1W = 2
    pool1H = 2
    pool1hStride = 2
    pool1vStride = 2
    # MaxPooling is a standard NDL node.
    pool1 = MaxPooling(conv1, pool1W, pool1H, pool1hStride, pool1vStride, imageLayout=$imageLayout$)

    # conv2
    kW2 = 5
    kH2 = 5
    cMap2 = 32
    hStride2 = 1
    vStride2 = 1
    # weight[cMap2, kW2 * kH2 * cMap1]
    # ConvNDReLULayer is defined in Macros.ndl
    conv2 = ConvNDReLULayer(pool1, kW2, kH2, cMap1, 400, cMap2, hStride2, vStride2, 10, 1)

    # pool2
    pool2W = 2
    pool2H = 2
    pool2hStride = 2
    pool2vStride = 2
    # MaxNDPooling is defined in Macros.ndl
    pool2 = MaxNDPooling(conv2, pool2W, pool2H, pool2hStride, pool2vStride, imageLayout=$imageLayout$)

    h1Dim = 128
    # DNNImageSigmoidLayer and DNNLayer are defined in Macros.ndl
    h1 = DNNImageSigmoidLayer(7, 7, cMap2, h1Dim, pool2, 1)
    ol = DNNLayer(h1Dim, labelDim, h1, 1)

    ce = CrossEntropyWithSoftmax(labels, ol)
    errs = ErrorPrediction(labels, ol)

    # Special Nodes
    FeatureNodes = (features)
    LabelNodes = (labels)
    CriterionNodes = (ce)
    EvalNodes = (errs)
    OutputNodes = (ol)
]

数据集：

可能是因为图片用的白纸上拍的，但白纸拍的也有阴影，清楚模糊的区别。总的来说这种图片数据可能还是好一些。error率为0%。

运行结果：