이번역페이지는최신내용을담고있지않습니다。최신내용을문으로보려면여기를클릭하십시오。
importKerasNetwork
사전훈련된Keras신경망과가중치가져오기
설명
예제
입력marketing수
출력marketing수
팁
importKerasNetwork
함수는다음과같은Keras계층유형을갖는신경망을가져올수있으며,몇가지제한사항이있습니다。신경망에그밖의다른유형의계층이포함된경우,오류메시지가반환됩니다。이경우에도importKerasLayers
를사용하여신경망아키텍처와가중치를가져올수있습니다。지원되는Keras계층 대응되는深度学习工具箱계층 添加
additionLayer
다음과같은활성화이름을갖는
激活
:“elu”
“relu”
“线性”
“softmax”
“乙状结肠”
的双曲正切
계층:
고급활성화:
ELU
Softmax
线性整流函数(Rectified Linear Unit)
LeakyReLU
PReLu
*
계층:
nnet.keras.layer.PreluLayer
AveragePooling2D
averagePooling2dLayer
BatchNormalization
batchNormalizationLayer
双向(LSTM (__))
bilstmLayer
连接
depthConcatenationLayer
Conv2D
convolution2dLayer
Conv2DTranspose
transposedConv2dLayer
CuDNNLSTM
lstmLayer
密集的
fullyConnectedLayer
DepthwiseConv2D
groupedConvolution2dLayer
辍学
dropoutLayer
嵌入
wordEmbeddingLayer
(文本分析工具箱)平
nnet.keras.layer.FlattenCStyleLayer
GlobalAveragePooling2D
globalAveragePooling2dLayer
GlobalMaxPooling2D
globalMaxPooling2dLayer
格勒乌
gruLayer
输入
imageInputLayer
LSTM
lstmLayer
MaxPooling2D
maxPooling2dLayer
乘
multiplicationLayer
SeparableConv2D
groupedConvolution2dLayer
또는convolution2dLayer
UpSampling2D
resize2dLayer
(图像处理工具箱)UpSampling3D
resize3dLayer
(图像处理工具箱)ZeroPadding2D
nnet.keras.layer.ZeroPadding2DLayer
*PReLU계층의경우,
importKerasNetwork
함수가벡터값스케일링파라미터를벡터소의평균값으로바꿉니다。가져온후에파라미터를다시벡터로변경할수있습니다。예제는导入Keras PReLU图层항목을참조하십시오。importKerasNetwork
함수는다음과같은Keras손실함수를지원합니다。mean_squared_error
categorical_crossentropy
sparse_categorical_crossentropy
binary_crossentropy
다중입력/다중출력(MIMO)을포함하는Keras신경망을가져올수있습니다。신경망이입력값에대한입력크기정보를포함하고출력값에대한손실정보를포함하는경우
importKerasNetwork
를사용하십시오。그밖의경우에는importKerasLayers
를사용하십시오。importKerasLayers
함수는입력값과출력값을위한자리@ @시자계층을삽입합니다。가져온후에는findPlaceholderLayers
를사용하여자리@ @시자계층을찾고replaceLayer
를사용하여자리@ @시자계층을바꿀수있습니다。米姆Keras신경망을가져오는워크플로는MIMO ONNX™신경망을가져오는워크플로와같습니다。예제는导入和组装多输出ONNX网络항목을참조하십시오。여러개의입력값과여러개의출력값을갖는딥러닝신경망에대해자세히알아보려면多输入多输出网络항목을참조하십시오。사전훈련된신경망을새영상에대한예측또는전이학습을위해사용하려면모델을훈련시킬때사용한영상의전처리방식과동일하게영상을전처리해야합니다。가장일반적인전처리연산으로영상크기조정하기,평균값영상빼기,영상을RGB에서BGR형식으로변환하기등을들수있습니다。
훈련및예측을위한상전처리에대한자세한내용은딥러닝을위해상전처리하기항목을참조하십시오。
호환성관련고려사항
참고 문헌
[1] Keras: Python深度学习库。https://keras.io.
참고 항목
exportONNXNetwork
|importCaffeLayers
|importCaffeNetwork
|importKerasLayers
|importONNXLayers
|importONNXNetwork