深度学习之图片转成pkl格式的训练集

原文地址：https://blog.csdn.net/csuzhaoqinghui/article/details/54346637

 

怎么使用python将文件夹下的图片转换pkl格式的训练集。

首先需要读取图片和图片所属的类别。

主要思路是：1.生成train.Thumbs.db 
                        2.生成label.Thumbs.db 
                        3.将train.Thumbs.db 和label.Thumbs.db合并成train.pkl
 

import os
import cPickle    
import numpy  
from PIL import Image
 
'''
###############################################################
'''
#flag='train'
flag='test'
filename='test'
outfile='gait'+flag+'.pkl'
imgsize=32
'''
################################################################
'''
 
#获得文件夹下所有图片
def getFilePicture(folder):
    assert os.path.exists(folder)
    assert os.path.isdir(folder)
    PictureList = os.listdir(folder)
    PictureList = [str(folder)+'/'+item for item in PictureList]
    return PictureList
 
#转换函数  
def writepkltemp(rootDir): 
    for lists in os.listdir(rootDir): 
        path =rootDir+'/'+lists
        if os.path.isdir(path): 
            writepkltemp(path) 
        else:
            label=path.split('/',-1)[-2]
            print path,label
            img = Image.open(path)
            img = img.resize((imgsize,imgsize))
            if img.size>2:
               img=img.convert("1")
            img_ndarray = numpy.asarray(img, dtype='float64')#/256
            #print img_ndarray.shape
            global vector
            global vector_label
            global num
            #print len(numpy.ndarray.flatten(img_ndarray))
            vector[num]=numpy.ndarray.flatten(img_ndarray)
            vector_label[num]=int(label)
            num=num+1
 
#保存pkl格式图片集            
def writepkl(filename):
    writepkltemp(filename)
 
#获得文件夹下所有图片的数量
def getnum(rootDir): 
    for lists in os.listdir(rootDir): 
        path =rootDir+'/'+lists
        if os.path.isdir(path): 
            getnum(path) 
        else:
            global n
            n=n+1
 
#图片总量            
def getimgnum(numpath):
    getnum(numpath)
    
if __name__ == "__main__":
    num=0
    n=0
    numpath=filename
    getimgnum(numpath) 
    vector=numpy.empty((n,imgsize*imgsize))
    vector_label=numpy.empty(n)
    writepkl(filename)
    vector_label=vector_label.astype(numpy.int)
    write_file=open(outfile,'wb')
    cPickle.dump([vector[0:n],vector_label[0:n]], write_file,-1)
    write_file.close()