何Caffe配置每层结构近刚电脑装Caffe由于神经中国络同层结构同类型层同参数所根据Caffe官中国说明文档做简单总结1.VisionLayers1.1卷积层(Convolution)类型:CONVOLUTION例layers{name:"conv1"type:CONVOLUTIONbottom:"data"top:"conv1"blobs_lr:1#learningratemultiplierforthefiltersblobs_lr:2#learningratemultiplierforthebiasesweight_decay:1#weightdecaymultiplierforthefiltersweight_decay:0#weightdecaymultiplierforthebiasesconvolution_param{num_output:96#learn96filterskernel_size:11#eachfilteris11x11stride:4#step4pixelsbetweeneachfilterapplicationweight_filler{type:"gaussian"#initializethefiltersfromaGaussianstd:0.01#distributionwithstdev0.01(defaultmean:0)}bias_filler{type:"constant"#initializethebiasestozero(0)value:0}}}blobs_lr:习率调整参数面例设置权重习率运行求解器给习率同偏置习率权重两倍
用符号:m表示,我们走两步的距离约是 米,
e on Novembe