Import a network saved in the ONNX format as a function and modify the network parameters.
Import the pretrainedSimpleNet3fc.onnx网络作为函数。SimpleNet3FCis a simple convolutional neural network trained on digit image data. For more information on how to create a network similar toSimpleNet3FC， 看创建简单的图像分类网络。
ImportSimpleNet3fc.onnx使用experconnxfunction，返回ONNXParameters包含网络参数的对象。该函数还会在包含网络体系结构的当前文件夹中创建一个新的模型函数。将模型函数的名称指定为simplenetFcn。
包含导入的ONNX网络的函数已保存到文件SimpleNetfcn.m。要了解如何使用此功能，请键入：帮助SimpleEnetFCN。
显示在培训期间更新的参数（params.learnables）and the parameters that remain unchanged during training (参数）。
ans =struct with fields:imageInput_mean：[1×1 dlarray] conv_w：[5×5×1×20 dlarray] conv_b：[20×1 dlarray] batchnorm_scale：[20×1 dlarray] batchnorm_b：[20×1 dlarray] fc_1_w：[24×24×24×20×20 dlarray] fc_1_b：[20×1 dlarray] fc_2_w：[1×1×1×20×20 dlarray] fc_2_b：[20×1 dlarray] fc_3_w：[1×1×1×1×20×10×10×10 dlarray]×1 dlarray]
ans =struct with fields:ConvStride1004: [2×1 dlarray] ConvDilationFactor1005: [2×1 dlarray] ConvPadding1006: [4×1 dlarray] ConvStride1007: [2×1 dlarray] ConvDilationFactor1008: [2×1 dlarray] ConvPadding1009: [4×1 dlarray] ConvStride1010:[2×1 dlarray] ConvDilationFactor1011: [2×1 dlarray] ConvPadding1012: [4×1 dlarray] ConvStride1013: [2×1 dlarray] ConvDilationFactor1014: [2×1 dlarray] ConvPadding1015: [4×1 dlarray]
The network has parameters that represent three fully connected layers. You can add a fully connected layer in the original parametersparams在两层之间fc_2andfc_3。The new layer might increase the classification accuracy.
To see the parameters of the convolutional layersfc_2andfc_3，打开模型功能simplenetFcn。
Scroll down to the layer definitions in the functionsimplenetFcn。The code below shows the definitions for layersfc_2andfc_3。
％cons：[weights, bias, stride, dilationFactor, padding, dataFormat, NumDims.fc_2] = prepareConvArgs(Vars.fc_2_W, Vars.fc_2_B, Vars.ConvStride1010, Vars.ConvDilationFactor1011, Vars.ConvPadding1012, 1, NumDims.fc_1, NumDims.fc_2_W); Vars.fc_2 = dlconv(Vars.fc_1, weights, bias,“大步”，，，，stride,'DilationFactor'，，，，dilationFactor,'Padding'，，，，padding,'DataFormat'，，，，dataFormat);％cons：[权重，偏见，大步，扩张量，填充，填充，dataFormat，numdims.fc_3] = PrepareConvargs（vars.fc_3_w，vars.fc_3_b，vars.convstride1013，vars.convdilationfactor1014;vars.fc_3 = dlconv（vars.fc_2，striges，bias，“大步”，，，，stride,'DilationFactor'，，，，dilationFactor,'Padding'，，，，padding,'DataFormat'，，，，dataFormat);
命名新层fc_4，，，，because each added parameter name must be unique. TheaddParameterfunction always adds a new parameter sequentially to theparams.learnablesor参数structure. The order of the layers in the model functionsimplenetFcndetermines the order in which the network layers are executed. The names and order of the parameters do not influence the execution order.
Add a new fully connected layerfc_4具有与fc_2。
显示更新的可学习和不可学习的参数。
ans =struct with fields:imageInput_mean：[1×1 dlarray] conv_w：[5×5×1×20 dlarray] conv_b：[20×1 dlarray] batchnorm_scale：[20×1 dlarray] batchnorm_b：[20×1 dlarray] fc_1_w：[24×24×24×20×20 dlarray] fc_1_b：[20×1 dlarray] fc_2_w：[1×1×1×20×20 dlarray] fc_2_b：[20×1 dlarray] fc_3_w：[1×1×1×1×20×10×10×10 dlarray]×1 dlarray] fc_4_w：[1×1×20×20 dlarray] fc_4_b：[20×1 dlarray]
ans =struct with fields:ConvStride1004: [2×1 dlarray] ConvDilationFactor1005: [2×1 dlarray] ConvPadding1006: [4×1 dlarray] ConvStride1007: [2×1 dlarray] ConvDilationFactor1008: [2×1 dlarray] ConvPadding1009: [4×1 dlarray] ConvStride1010:[2×1 dlarray] ConvdilationFactor1011：[2×1 dlarray] Convpadding1012：[4×1 dlarray] Convstride1013：[2×1 dlarray] ConvdilationFactor1014：[2×1 Dlarray] Corvpadding1115：[4×1 Dlarray：[4×1 Dlarray] fc_4;×1 dlarray] fc_4_dilationFactor：[2×1 dlarray] fc_4_padding：[4×1 dlarray]
修改模型功能的体系结构以反映params因此，您可以将网络使用新参数进行预测或重新训练网络。打开模型功能simplenetFcn。然后，添加完全连接的层fc_4在两层之间fc_2andfc_3，并更改卷积操作的输入数据dlconv对于层fc_3至Vars.fc_4。
The code below shows the new layerfc_4在其位置和层fc_2andfc_3。
％cons：[weights, bias, stride, dilationFactor, padding, dataFormat, NumDims.fc_2] = prepareConvArgs(Vars.fc_2_W, Vars.fc_2_B, Vars.ConvStride1010, Vars.ConvDilationFactor1011, Vars.ConvPadding1012, 1, NumDims.fc_1, NumDims.fc_2_W); Vars.fc_2 = dlconv(Vars.fc_1, weights, bias,“大步”，，，，stride,'DilationFactor'，，，，dilationFactor,'Padding'，，，，padding,'DataFormat'，，，，dataFormat);% Conv[weights, bias, stride, dilationFactor, padding, dataFormat, NumDims.fc_4] = prepareConvArgs(Vars.fc_4_W, Vars.fc_4_B, Vars.fc_4_Stride, Vars.fc_4_DilationFactor, Vars.fc_4_Padding, 1, NumDims.fc_2, NumDims.fc_4_W); Vars.fc_4 = dlconv(Vars.fc_2, weights, bias,“大步”，，，，stride,'DilationFactor'，，，，dilationFactor,'Padding'，，，，padding,'DataFormat'，，，，dataFormat);％cons：[权重，偏见，大步，扩张性，填充，填充，dataFormat，numdims.fc_3] = PrepareConvargs（vars.fc_3_w，vars.fc_3_b，vars.convstride1013，vars.convdilationfactor1014;vars.fc_3 = dlconv（vars.fc_4，strige，bias，“大步”，，，，stride,'DilationFactor'，，，，dilationFactor,'Padding'，，，，padding,'DataFormat'，，，，dataFormat);
