Lab-VOC数据集(多分类)制作
1.使用精灵标记助手标注
- 标记时对一张图片标记四次
- 导出时一张原图对应四张标记图
- 处理标记图时预先将标记图分为四类,对应每一分类,文件数量相同且与原图数量相等(这一步手动操作比较麻烦)
我的文件夹设置
转换灰度
import cv2
import os
path = 'attachments/0'
path1 = 'attachments/0_puts'
file_list = os.listdir(path)
for file in file_list:
I = cv2.imread(path+'/'+file)
imGray = cv2.cvtColor(I, cv2.COLOR_BGR2GRAY)
cv2.imwrite(path1+"/"+file,imGray)
cv2.imshow('frame',imGray)
cv2.waitKey(1)
灰度设值
- 我把四类分别设置成了固定值10,11,12,13
- 运行四次,每次更改对应路径
import cv2 as cv
import os
path = 'attachments/3_puts'
path1 = 'attachments/3_puts_13'
file_list = os.listdir(path)
for file in file_list:
img=cv.imread(path+"/"+file)
imgCopy = img
imgCopy[imgCopy > 0] = 13
cv.imwrite(path1+"/"+file,imgCopy)
四图合成
import os
import cv2
import numpy as np
def fourPicCompose(img0,img1,img2,img3):
imgResult = np.zeros((512,512)) # 已知图片大小
imgResult = img0 + img1 + img2 + img3
imgResult[imgResult > 20] = 10 # 防止重合,重合时值设置为第一类的值
return imgResult
path = 'attachments/0_puts_10'
file_list = os.listdir(path)
for file in file_list:
fileName=file[:-5] # 得到文件的基础名'(?)_'
print(fileName)
#读取四个文件块的特定名称的图片,
img0=cv2.imread('attachments/'+'0_puts_10'+'/'+fileName+'1.png',cv2.IMREAD_GRAYSCALE)
img1=cv2.imread('attachments/'+'1_puts_11'+'/'+fileName+'2.png',cv2.IMREAD_GRAYSCALE)
img2=cv2.imread('attachments/'+'2_puts_12'+'/'+fileName+'3.png',cv2.IMREAD_GRAYSCALE)
img3=cv2.imread('attachments/'+'3_puts_13'+'/'+fileName+'4.png',cv2.IMREAD_GRAYSCALE)
# 得到四合一合成图
imgResult=fourPicCompose(img0,img1,img2,img3)
cv2.imshow('frame',imgResult)
cv2.waitKey(1)
cv2.imwrite('attachments/results/'+fileName+'.png',imgResult)
Tips
- 无论用cv2还是PIL转换灰度图,直接读取后图片的shpae还是[???,???,3]。使用cv2读取时加上参数得cv2.imread(???,cv2.IMREAD_GRAYSCALE)