python opencv load image and splite train/test set

from sklearn import cross_validation
import cv2

def load_image_data():
    dir_path=r'E:\'
    # class_1_path=dir_path+r'\no'
    # class_2_path=dir_path+r'\other'
    # class_3_path=dir_path+r'\yes'
    X=[]
    Y=[]
    testX=[]
    testY=[]

    #print(class_1_path)
    filepathvec = []
    for rt, dirs, files in os.walk(dir_path):  # =pathDir
        # print('rt:',rt)
        # print('dirs:',dirs)
        for dir in dirs:
            # print(dir)
            # if dir.find('-')<0:
            #     continue
            for rt1, dirs, filenames in os.walk(rt+'\\'+dir):
                file_count=0
                for filename in filenames:
                    file_count+=1
                    if file_count>200:
                        break
                    # print (filename)
                    if filename.find('.') >= 0:
                        (shotname, extension) = os.path.splitext(filename)
                        # print shotname,extension
                        if extension == '.png' or extension == '.tif':
                            filepathvec.append(os.path.join('%s\\%s' % (rt+'\\'+dir, filename)))
                            # print (filename)
        break
                    
    # exit()
    print(len(filepathvec))
    for file in filepathvec:
        # if len(X)>3200:
        #     break
        print(file)
        img=cv2.imread(file,0)
        # print(np.shape(img))
        #cv2.namedWindow("the window")
        #cv2.imshow("the window",img)
        image_width,image_height=np.shape(img)
        print(image_width)
        fzoom=20.0/float(image_width)
        img=cv2.resize(img,(int(float(image_height)*fzoom),20), interpolation=cv2.INTER_CUBIC)

        img=cv2.rotate(img,cv2.ROTATE_90_CLOCKWISE)
        img=255-img #反色处理
        # cv2.imshow('xx', img)
        # cv2.waitKey()

        #print(img)
        #print(len(img))
        #print(np.shape(img))
        X.append(img)
        Y.append(get_label(file))

        #cv2.namedWindow("the window")
        #cv2.imshow("the window",img)
        #break
        #多块平移提取处理
        # img_src_height=np.shape(img)[0]
        # img_src_width=np.shape(img)[1]
        # for i in range(1,3):
        #     img_crop=np.array(img[i:-1][i:-1])
        #     img_crop = cv2.resize(img_crop, (28, 28), interpolation=cv2.INTER_CUBIC)
        #     img_crop = 255 - img_crop  # 反色处理
        #     X.append(img_crop)
        #     Y.append(get_label(file))
        #
        # for i in range(1,3):
        #     img_crop = img[0:i][0:i]
        #     img_crop = cv2.resize(img_crop, (28, 28), interpolation=cv2.INTER_CUBIC)
        #     img_crop = 255 - img_crop  # 反色处理
        #     X.append(img_crop)
        #     Y.append(get_label(file))

    # print(X[0:5])
    # print(Y[0:5])
    print(len(X),len(Y))
    X=np.array(X)
    Y=np.array(Y)
    # testX=X
    # testY=Y

    X, testX, Y, testY = cross_validation.train_test_split(X, Y, test_size=0.2)
    print('train set:',len(X))
    print('test set:',len(testX))

    return X,Y,testX,testY