我們把需要做的資料放在Caffe_DataMaker 資料夾中，並且將資料按比例分為測試集資料夾train和訓練集資料夾val；在測試集資料夾中按照類別放在相應的資料夾中，資料夾命名為0,1，…..（從0開始命名），在測試集資料夾中進行同樣的操作，使用的python程式碼為：

# -*- coding: utf-8 -*-
"""
使用python製作用於caffe分類的lmdb資料來源
"""
import os
import io
def caffe_input_txt_maker(data_folder,outfile_name, phase = 'train'):
    # 計數檔案個數
 

    file_cnt = 0
    class_cnt = 0
    with io.open(outfile_name, 'wb+') as fobj:
        for folder_name in os.listdir(data_folder):
            label = folder_name.split('__')[0]
            folder_path = os.path.join(data_folder, folder_name)
            class_cnt += 1
            for file_name in os.listdir(folder_path):
                file_cnt += 1
 
  # 將資料夾名稱也新增入內
                if phase == 'train' :
                    file_path = folder_name + '/' + file_name
                if phase == 'test' :
                    file_path = file_name
                fobj.writelines( file_path +" "+str(label)+'\n')

    file_dir, base_name = os.path.split(outfile_name)
    file_name, ext = os.path.splitext(base_name)

    #new_outfile_name = file_dir + '/' + file_name + '_%d_%d' % (class_cnt, file_cnt) + ext
 

    #if os.path.exists(new_outfile_name): os.remove(new_outfile_name)
    #os.rename(outfile_name, new_outfile_name)
    print ('Done')

if __name__ == "__main__":
    caffe_input_txt_maker(data_folder = '/home/pcb/caffe/examples/Caffe_DataMaker/train',
                          outfile_name = "/home/pcb/caffe/examples/Caffe_DataMaker/train.txt", phase = 'train')
    caffe_input_txt_maker(data_folder = '/home/pcb/caffe/examples/Caffe_DataMaker/val',
                          outfile_name = "/home/pcb/caffe/examples/Caffe_DataMaker/val.txt", phase = 'test')

data_folder是存放訓練集的資料夾，outfile_name是輸出的train.txt,其內容如下：

val.txt的內容如下：

注意：待生成val.txt後，訓練集val資料夾下的圖片全部從類別資料夾中移動到val資料夾下，因為val.txt中圖片路徑前面沒有子目錄。
接下來就要寫create_imagenet.sh裡面的東西了，如下所示：

這#!/usr/bin/env sh
# Create the imagenet lmdb inputs
# N.B. set the path to the imagenet train + val data dirs
set -e

EXAMPLE=/home/pcb/caffe/examples/Caffe_DataMaker  #放置相應的資料夾
DATA=/home/pcb/caffe/examples/Caffe_DataMaker     #放置相應的資料夾
TOOLS=/home/pcb/caffe/build/tools

TRAIN_DATA_ROOT=/home/pcb/caffe/examples/Caffe_DataMaker/train/  #存放用於訓練的資料
VAL_DATA_ROOT=/home/pcb/caffe/examples/Caffe_DataMaker/val/      #存放用於測試的資料


# Set RESIZE=true to resize the images to 256x256. Leave as false if images have
# already been resized using another tool.
RESIZE=false
if $RESIZE; then
  RESIZE_HEIGHT=256
  RESIZE_WIDTH=256
else
  RESIZE_HEIGHT=0
  RESIZE_WIDTH=0
fi

if [ ! -d "$TRAIN_DATA_ROOT" ]; then
  echo "Error: TRAIN_DATA_ROOT is not a path to a directory: $TRAIN_DATA_ROOT"
  echo "Set the TRAIN_DATA_ROOT variable in create_imagenet.sh to the path" \
       "where the ImageNet training data is stored."
  exit 1
fi

if [ ! -d "$VAL_DATA_ROOT" ]; then
  echo "Error: VAL_DATA_ROOT is not a path to a directory: $VAL_DATA_ROOT"
  echo "Set the VAL_DATA_ROOT variable in create_imagenet.sh to the path" \
       "where the ImageNet validation data is stored."
  exit 1
fi

echo "Creating train lmdb..."

GLOG_logtostderr=1 $TOOLS/convert_imageset \
    --resize_height=$RESIZE_HEIGHT \
    --resize_width=$RESIZE_WIDTH \
    --shuffle \
    $TRAIN_DATA_ROOT \
    $DATA/train.txt \
    $EXAMPLE/train_lmdb

echo "Creating val lmdb..."

GLOG_logtostderr=1 $TOOLS/convert_imageset \
    --resize_height=$RESIZE_HEIGHT \
    --resize_width=$RESIZE_WIDTH \
    --shuffle \
    $VAL_DATA_ROOT \
    $DATA/val.txt \
    $EXAMPLE/val_lmdb

echo "Done."裡寫程式碼片

寫完之後在終端的目錄調整到Caffe_DataMaker下，使用命令sh create_imagenet.sh就會在Caffe_DataMaker資料夾下生成train_lmdb和val_lmdb,這樣就大功告成了！