如何忽略caffe中的日志打印未使用的图层 [英] How to ignore log print a unused layer in caffe

问题描述

我有一个原型，如下:

layer {
  name: "data"
  type: "HDF5Data"
  top: "data"
  top: "label"
  include {
    phase: TRAIN
  }
  hdf5_data_param {
    source: "./train.txt"
    batch_size: 2
  }
}
layer {
  name: "data_scale2"
  type: "HDF5Data"
  top: "data_scale2"
  top: "label_scale2"
  include {
    phase: TRAIN
  }
  hdf5_data_param {
    source: "./train_scale2.txt"
    batch_size: 2
  }
}

其中第二层包含原始数据2的比例因子，我将其命名为data_scale2.在训练期间，我只使用了data，label和data_scale2，而没有使用label_scale2.因此，当我进行训练时，label_scale2的值总是在我的终端上打印.我该如何忽略呢?该日志看起来像

where the second layer contains the scale factor of 2 of the original data, I named it as data_scale2. During training, I just used the data,label and data_scale2 without using label_scale2. Hence, when I run training, the value of label_scale2 always printing in my terminal. How can I ignore this? This log looks like

I0302 18:01:57.356312 30995 solver.cpp:245]     Train net output #221177: label_scale2 = 0
I0302 18:01:57.356314 30995 solver.cpp:245]     Train net output #221178: label_scale2 = 0
I0302 18:01:57.356320 30995 solver.cpp:245]     Train net output #221179: label_scale2 = 0
I0302 18:01:57.356324 30995 solver.cpp:245]     Train net output #221180: label_scale2 = 0
I0302 18:01:57.356328 30995 solver.cpp:245]     Train net output #221181: label_scale2 = 0
I0302 18:01:57.356329 30995 solver.cpp:245]     Train net output #221182: label_scale2 = 0
I0302 18:01:57.356333 30995 solver.cpp:245]     Train net output #221183: label_scale2 = 0

推荐答案

常规解决方案:
如果您想使"top"层静音，则可以使用 "Silence" 层:

General solution:
If you wan to silence a "top" layer you can use "Silence" layer:

layer {
  type: "Silence"
  name: "silence_this_layer_for_me"
  bottom: "label_scale2"
}

就是这样！您将不会再听到label_scale2的声音

And that's it! You won;t hear from label_scale2 again

HDF5Data解决方案
另外，您不必公开hdf5输入的所有datasets.您可以注释多余的"top":

A solution for HDF5Data
Alternatively, you do not have to expose all datasets of an hdf5 input. You can simply comment out the redundant "top":

layer {
  name: "data_scale2"
  type: "HDF5Data"
  top: "data_scale2"
  # top: "label_scale2" 
  include {
    phase: TRAIN
  }
  hdf5_data_param {
    source: "./train_scale2.txt"
    batch_size: 2
  }
}

这篇关于如何忽略caffe中的日志打印未使用的图层的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！