caffe之均值文件

sudo python convert_mean.py mean.binaryproto mean.npy

一.二进制的均值计算

caffe中使用的均值数据格式是binaryproto, caffe为我们提供了一个计算均值的文件compute_image_mean.cpp, 在caffe目录下的build/tool文件中

sudo build/tools/compute_image_mean examples/mnist/mnist_train_lmdb examples/mnist/mean.binaryproto

第一个参数：examples/mnist/mnist_train_lmdb， 表示需要计算均值的数据，格式为lmdb的训练数据

第二个参数：examples/mnist/mean.binaryproto， 计算出来的结果保存文件。

二. python格式的均值计算

由python脚本实现

  1 #!/usr/bin/env python                                                       
  2 import numpy as np
  3 import sys,caffe
  4 
  5 if len(sys.argv)!=3:
  6     print "Usage: python convert_mean.py mean.binaryproto mean.npy"
  7     sys.exit()
  8 
  9 blob = caffe.proto.caffe_pb2.BlobProto()
 10 bin_mean = open( sys.argv[1] , 'rb' ).read()
 11 blob.ParseFromString(bin_mean)
 12 arr = np.array( caffe.io.blobproto_to_array(blob) )
 13 npy_mean = arr[0]
 14 np.save( sys.argv[2] , npy_mean )

将这个脚本保存为convert_mean.py，调用

python convert_mean.py mean.binaryproto mean.npy

convert_mean.py  上述脚本文件

mean.binaryproto 上述的二进制文件

mean.npy 为保存的python格式的均值文件

自己使用过的一个脚本

  1 #!usr/bin/env python
  2 import PIL
  3 import image
  4 import sys
  5 import time
  6 import os
  7 import numpy as np
  8 from matplotlib import pyplot as plt
  9 
 10 start = time.time()                                                         
 11 
 12 # Make sure that caffe is on the python path
 13 caffe_root = '/home/caffe/'  
 14 sys.path.insert(0, caffe_root + 'python')
 15 
 16 import caffe
 17 # "source" is the binary file converted by the command shell 
 18 # "des" is the binary file with python format converted from "source"
 19 source = caffe_root + 'examples/mnist/mean.binaryproto'
 20 des = caffe_root + 'examples/mnist/mean_1.npy'
 21 
 22 # BlobProto object
 23 blob = caffe.proto.caffe_pb2.BlobProto()
 24 data = open( source , 'rb' ).read()
 25 # parsing source data
 26 blob.ParseFromString(data)
 27 # convert to npy format
 28 arr = np.array( caffe.io.blobproto_to_array(blob) )
 29 out = arr[0]
 30 # save the converted result
 31 np.save( des , out )
~                         

注：

sudo pip install image

权限不够，获取root权限

在进入python编辑模式时import caffe并没有报错，运行脚本文件时却提示我没有 module caffe

sudo python convert_mean.py mean.binaryproto mean.npy

将sudo 去掉后可以运行

本章主要参考http://www.cnblogs.com/denny402/p/5102328.html