caffe digits环境下finetune时prototxt文件修改方法

1、删除两个data层中的多余参数，剩余如下形式：

layer {
  name: "data"
  type: "Data"
  top: "data"
  top: "label"
  include {
    phase: TRAIN
  }
  transform_param {
    mirror: true
    crop_size: 227
  }
}

layer {
  name: "data"
  type: "Data"
  top: "data"
  top: "label"
  include {
    phase: TEST
  }
  transform_param {
    mirror: false
    crop_size: 227
  }
}

2、将所有include {phase: TRAIN}替换成include {stage: "train"}
将所有include {phase: TEST}替换成include {stage: "val"}
在loss层添加exclude {stage: "deploy"}

3、将fc8层的名字替换成fc8-re（只要换一个名字即可，因为分类数量与pretrained model数量不同时会初始化错误），并删除fc8中的num_output行

4、在文件最后添加softmax层

layer {
  name: "prob"
  type: "Softmax"
  bottom: "fc8"
  top: "prob"
  include{
      stage:"deploy"
      }
}