2020年4月15日更新：
原代码中图片都是jpg的，更新了可支持.png,.bmp,.jpg不同格式混合在一起的情况。
完整demo上传百度盘了，有兴趣可以看一下。
链接: https://pan.baidu.com/s/1mgqUsJmn6ynFHqKVAmYPSA 提取码: xgnt
a.gif
（上面git图打不开的话请看连接，这不知道算不算简书网页的"bug"，在文章编辑界面轻易能显示动态gif，发布出去就难加载出来-_-!）
简介：本篇文章把darknet训练YOLO模型的过程中需要用到的一些预处理函数打包封装成一个类，方便应用，图片标注工具是Labelme。封装的类中包含labelme标注的标签文件转为darknetYOLO所需的txt格式；数据增强（包括水平、竖直翻转，平滑滤波，gamma变换），在数据增强的时候也会把对应的txt文件做相同的处理；从json标注文件复查标签；从txt文件复查标签。
造轮子。
1.darknet训练的YOLO数据集应该如下图这样，图片和标签文件放在同一个文件夹，且图片和标签文件名一样。
image.png
2. txt标签格式
<class> <x_center> <y_center> <width> <height>
由于图片可能有尺度缩放所以YOLO里边的位置用的是图片的百分比坐标,如下
image.png
3.labelme标注的json文件转txt格式
labelme标注的文件是json格式，保存的坐标是绝对坐标，需要转换为2.中的txt标签格式。
参考：https://www.jianshu.com/p/61436306b479
4.示例：
4.1假设已经标注好了图片，且图片和标签文件保存如下：原始图片保存在srcImg，标注的json文件保存在jsonLabel。
image.png
image.png
4.2 把json标注格式转为txt格式，并加一些数据增强操作的完整流程，类代码有点长附在文末。
if __name__ == "__main__":
srcImageFolder = "K:\imageData\polarity\image1"# 存放原始图片的文件夹,有标注过的和没标注过的图片
jsonFolder = "K:\imageData\polarity\jsonLabel1"# labelme标注的json文件保存的文件夹
labeledImageFolder = "K:/imageData/polarity/temp/labeledImage"# 筛选出所有标注过的图片
txtFolder = "K:/imageData/polarity/temp/txtLabel"# txt文件保存的路径
objFolder = "K:/imageData/polarity/temp/obj"# 把图片和对应的txt文件放在一起，可以直接拿去训练的那种
trainAndTestFileSaveFolder = "K:/imageData/polarity/temp"
# 创建文件夹
try:
if not os.path.exists(labeledImageFolder):
os.makedirs(labeledImageFolder)
if not os.path.exists(txtFolder):
os.makedirs(txtFolder)
if not os.path.exists(objFolder):
os.makedirs(objFolder)
except:
print("os.makedirs failed!")
# 初始化类
tool = darknetTool()
# 获取所有原始标注文件
jsonPathList = tool.getAllJsonFilePath(jsonFolder)
# 从json标签文件复查标注
# for item in jsonPathList:
# tool.showLabelFromJson(item,imageFolder)
# 从json标注文件查找图片，并保存到labeledImageFolder
srcLabeledImagePathList = tool.fromJsonFileToSearchImage(jsonPathList, srcImageFolder)
tool.copyAllLabeledImageToNewFolder(srcLabeledImagePathList, labeledImageFolder)
# json格式转为txt格式
for item in jsonPathList:
tool.changeLabelFromJsonToTxt(item, txtFolder)
txtPathList = tool.getAllTxtPath(txtFolder)
# 获取在labeledImageFolder中的图片的路径
dstLabeledImagePathList = tool.getAllImagePath(labeledImageFolder)
# 从txt标签文件查看标注结果
# 所有标注的图片路径都在dstLabeledImagePathList，通过图片路径去找对应的txt文件
# 所有的txt文件都在txtFolder文件夹下
# for imagePath in dstLabeledImagePathList:
# tool.showLabelFromTxt(imagePath,txtFolder)
# 数据增强
# 均值滤波
for imagePath in dstLabeledImagePathList:
tool.imageAugment_smooth(imagePath, txtFolder, labeledImageFolder,n=3) # n=9是滤波核大小
# 水平翻转
for imagePath in dstLabeledImagePathList:
tool.imageAugment_flip(imagePath, txtFolder, labeledImageFolder, flag=0) # flag=0 水平翻转
# 竖直翻转
for imagePath in dstLabeledImagePathList:
tool.imageAugment_flip(imagePath, txtFolder, labeledImageFolder, flag=1) # flag=1 竖直翻转
# 中心对称
for imagePath in dstLabeledImagePathList:
tool.imageAugment_flip(imagePath, txtFolder, labeledImageFolder, flag=2) # flag=2 中心对称
# gamma变换
for imagePath in dstLabeledImagePathList:
tool.imageAugment_gamma(imagePath, txtFolder, labeledImageFolder, gamma=0.5)
tool.imageAugment_gamma(imagePath, txtFolder, labeledImageFolder, gamma=0.8)
tool.imageAugment_gamma(imagePath, txtFolder, labeledImageFolder, gamma=1.5)
#tool.imageAugment_gamma(imagePath, txtFolder, labeledImageFolder, gamma=2)
#图片重命名
# 图片和标签重命名为00001.jpg 00001.txt.......（可选）
# n=5表示zfill（）的位数，n=5则为00001.jpg...，n=3则为001.jpg....
tool.renameAllLabeledImageAndTxt(txtFolder, labeledImageFolder, n=6, startN=0)
#检查转换的最终结果
dstLabeledImagePathList = tool.getAllImagePath(labeledImageFolder)
for imagePath in dstLabeledImagePathList:
tool.showLabelFromTxt(imagePath,txtFolder)
#划分训练集和测试集
imagePathList = tool.getAllImagePath(labeledImageFolder)
# prePath = "data/obj/"则train.txt文件中的路径就会是data/obj/00001.jpg,data/obj/00002.jpg......
# prop=0.8表示80%的数据用来训练，其余的做验证集
tool.devideTrainSetAndTestSet(imagePathList, trainAndTestFileSaveFolder, prePath="data/obj/", prop=0.8)
4.3结果
image.png
- labeledImage文件夹保存图片
-
txtLabel文件夹保存TXT格式的标签文件
image.png
- train.txt文件包含需要训练的图片的路径
-
test.txt文件包含验证集的图片的路径
image.png
4.4复查结果
#从txt文件复查数据增强后的结果
txtPathList_2 = tool.getAllTxtPath(txtFolder)
for item in txtPathList_2:
tool.showLabelFromTxt(item,labeledImageFolder)
image.png
5.总结
这个类就是一些函数的封装，单独拿去用也是没有什么问题的。
6.代码
2021年8月2日更新重命名填充位数，保证一定可以重命名
import os
import numpy as np
import cv2
import shutil
import json
import random
class darknetTool():
def __init__(self, help_=False):
if help_:
self.getHelp()
def getAllImagePath(self, folder, recursion=False):
expandName = ["jpg", "JPG", "jpeg", "JPEG", "png", "PNG", "bmp", "BMP"]
imagePathList = []
if not recursion:
for item in os.listdir(folder):
if os.path.isdir(os.path.join(folder, item)):
continue
if item.rsplit(".",1)[-1] in expandName:
imagePathList.append(os.path.join(folder, item))
return imagePathList
else:
for item in os.listdir(folder):
if os.path.isdir(os.path.join(folder, item)):
subPathList = self.getAllImagePath(os.path.join(folder, item), True)
imagePathList.extend(item for item in subPathList)
else:
if item.rsplit(".",1)[-1] in expandName:
imagePathList.append(os.path.join(folder, item))
return imagePathList
def getAllJsonFilePath(self, folder):
jsonPathList = []
for item in os.listdir(folder):
if item.rsplit(".",1)[-1] == 'json':
jsonPathList.append(os.path.join(folder, item))
return jsonPathList
def getAllTxtPath(self, folder):
txtPathList = []
for item in os.listdir(folder):
if item.rsplit(".",1)[-1] == 'txt':
txtPathList.append(os.path.join(folder, item))
return txtPathList
def fromJsonFileToSearchImage(self, jsonPathList, imageFolder):
imagePathList = []
for item in jsonPathList:
with open(item, "r", encoding='utf-8') as f:
jsonData = json.load(f)
imageName = jsonData["imagePath"].rsplit("\\",1)[-1]
jsonName = item.rsplit("\\",1)[-1]
imagePath = os.path.join(imageFolder, imageName)
# print(imagePath)
imagePathList.append(imagePath)
return imagePathList
def copyAllLabeledImageToNewFolder(self, imagePathList, saveFolder):
for item in imagePathList:
imageName = item.rsplit("\\",1)[-1]
savePath = os.path.join(saveFolder, imageName)
print("copyAllLabeledImageToNewFolder: srcPath", item)
print("copyAllLabeledImageToNewFolder: dstPath", savePath)
shutil.copyfile(item, savePath)
def renameAllLabeledImageAndTxt(self, txtFolder, labeledImageFolder, n=5, startN=0):
# extendName是图像后缀名，n用于zfill(n)，00001.jpg
i = startN
imagePathList = self.getAllImagePath(labeledImageFolder)
if (len(imagePathList[0].split("\\")[-1])-4) == n:
n += 1
for imagePath in imagePathList:
i += 1
imageName_dst = str(i).zfill(n) + "." + imagePath.rsplit(".",1)[-1]
image_srcPath = imagePath
image_dstPath = os.path.join(labeledImageFolder, imageName_dst)
txtName_src = imagePath.rsplit("\\",1)[-1].rsplit(".",1)[0] +".txt"
txtName_dst = str(i).zfill(n) + ".txt"
txt_srcPath = os.path.join(txtFolder, txtName_src)
txt_dstPath = os.path.join(txtFolder, txtName_dst)
try:
if not os.path.exists(image_dstPath):
os.rename(image_srcPath, image_dstPath)
if not os.path.exists(txt_dstPath):
os.rename(txt_srcPath, txt_dstPath)
print("renameAllLabeledImageAndTxt")
print("image_srcPath:", image_srcPath)
print("image_dstPath:", image_dstPath)
print("txt_srcPath:", txt_srcPath)
print("txt_dstPath:", txt_dstPath)
except:
print("ERROR:renameAllLabeledImageAndTxt error")
print("image_srcPath:", image_srcPath)
print("image_dstPath:", image_dstPath)
print("txt_srcPath:", txt_srcPath)
print("txt_dstPath:", txt_dstPath)
def changeLabelFromJsonToTxt(self, jsonPath, txtSaveFolder):
with open(jsonPath, "r", encoding='utf-8') as f:
jsonData = json.load(f)
img_h = jsonData["imageHeight"]
img_w = jsonData["imageWidth"]
txtName = jsonPath.rsplit("\\",1)[-1].rsplit(".",1)[0] + ".txt"
txtPath = os.path.join(txtSaveFolder, txtName)
with open(txtPath, "w") as f:
for item in jsonData["shapes"]:
label = item["label"]
pt1 = item["points"][0]
pt2 = item["points"][1]
xCenter = (pt1[0] + pt2[0]) / 2
yCenter = (pt1[1] + pt2[1]) / 2
obj_h = pt2[1] - pt1[1]
obj_w = pt2[0] - pt1[0]
f.write(" {} ".format(label))
f.write(" {} ".format(xCenter / img_w))
f.write(" {} ".format(yCenter / img_h))
f.write(" {} ".format(obj_w / img_w))
f.write(" {} ".format(obj_h / img_h))
f.write(" \n")
def showLabelFromJson(self, jsonPath, imageFolder):
cv2.namedWindow("img", 0)
font = cv2.FONT_HERSHEY_SIMPLEX
with open(jsonPath, "r") as f:
jsonData = json.load(f)
imageName = jsonData["imagePath"].rsplit("\\",1)[-1]
imagePath = os.path.join(imageFolder, imageName)
img = cv2.imread(imagePath)
print("showLabelFromJson: image path:", imagePath)
for item in jsonData["shapes"]:
label = item["label"]
p1 = (int(item["points"][0][0]), int(item["points"][0][1]))
p2 = (int(item["points"][1][0]), int(item["points"][1][1]))
cv2.putText(img, label, (p1[0], p1[1] - 10), font, 1.2, (0, 0, 255), 2)
cv2.rectangle(img, p1, p2, (0, 255, 0), 2)
cv2.imshow("img", img)
cv2.waitKey(0)
cv2.destroyWindow("img")
def showLabelFromTxt(self, imagePath, txtFolder):
# label xcenter ycenter w h
txtName = imagePath.rsplit("\\",1)[-1].rsplit(".",1)[0] + ".txt"
txtPath = os.path.join(txtFolder, txtName)
print("showLabelFromTxt: image path:", imagePath)
img = cv2.imread(imagePath)
h, w = img.shape[:2]
cv2.namedWindow("img", 0)
font = cv2.FONT_HERSHEY_SIMPLEX
with open(txtPath, "r") as f:
lines = f.readlines()
for line in lines:
tempL = line.split(" ")
label = tempL[1]
obj_w = float(tempL[7])
obj_h = float(tempL[9])
topLeftx = (float(tempL[3]) - obj_w / 2) * w
topLefty = (float(tempL[5]) - obj_h / 2) * h
bottomRightx = (float(tempL[3]) + obj_w / 2) * w
bottomRighty = (float(tempL[5]) + obj_h / 2) * h
p1 = (int(topLeftx), int(topLefty))
p2 = (int(bottomRightx), int(bottomRighty))
cv2.putText(img, label, (p1[0], p1[1] - 10), font, 1.2, (0, 255, 255), 2)
cv2.rectangle(img, p1, p2, (0, 255, 0), 2)
cv2.imshow("img", img)
cv2.waitKey(0)
cv2.destroyWindow("img")
def imageAugment_smooth(self, imagePath, txtFolder, labeledImageFolder,n=5):
txtName = imagePath.rsplit("\\",1)[-1].rsplit(".",1)[0] + ".txt"
txtPath = os.path.join(txtFolder, txtName)
print("smooth: image path:", imagePath)
img = cv2.imread(imagePath)
imageName_smooth = txtPath.rsplit("\\",1)[-1].split(".txt")[0] + "_smooth." + imagePath.rsplit(".",1)[-1]
imagePath_smooth = os.path.join(labeledImageFolder, imageName_smooth)#平滑图像保存路径
txtName_smooth = txtPath.rsplit("\\",1)[-1].split(".txt")[0] + "_smooth.txt"
txtPath_smooth = os.path.join(txtFolder, txtName_smooth)#标签保存路径
shutil.copyfile(txtPath, txtPath_smooth)
print("imageAugment_smooth: source txtPath", txtPath)
print("imageAugment_smooth: dst txtPath", txtPath_smooth)
print("imageAugment_smooth: source imagePath", imagePath)
print("imageAugment_smooth: dst imagePath", imagePath_smooth)
dst = cv2.blur(img, (n, n))
cv2.imwrite(imagePath_smooth, dst)
def imageAugment_flip(self, imagePath, txtFolder, labeledImageFolder,flag=0):
# flag = 0水平翻转 flag = 1，竖直翻转, flag=2 水平翻转+竖直翻转
txtName = imagePath.rsplit("\\",1)[-1].rsplit(".",1)[0] + ".txt"
txtPath = os.path.join(txtFolder, txtName)
print("imageAugment_flip: image path:", imagePath)
img = cv2.imread(imagePath)
if flag == 0:
imageName_flip = imagePath.rsplit("\\",1)[-1].rsplit(".",1)[0] + "_flipx." + imagePath.rsplit("\\",1)[-1].rsplit(".",1)[-1]
imagePath_flip = os.path.join(labeledImageFolder, imageName_flip)
txtName_flip = txtPath.rsplit("\\",1)[-1].split(".txt")[0] + "_flipx.txt"
txtPath_flip = os.path.join(txtPath.rsplit("\\",1)[0:-1][0], txtName_flip)
elif flag == 1:
imageName_flip = imagePath.rsplit("\\",1)[-1].rsplit(".",1)[0] + "_flipy." + imagePath.rsplit("\\",1)[-1].rsplit(".",1)[-1]
imagePath_flip = os.path.join(labeledImageFolder, imageName_flip)
txtName_flip = txtPath.rsplit("\\",1)[-1].split(".txt")[0] + "_flipy.txt"
txtPath_flip = os.path.join(txtPath.rsplit("\\",1)[0:-1][0], txtName_flip)
elif flag == 2:
imageName_flip = imagePath.rsplit("\\",1)[-1].rsplit(".",1)[0] + "_flipxy." + imagePath.rsplit("\\",1)[-1].rsplit(".",1)[-1]
imagePath_flip = os.path.join(labeledImageFolder, imageName_flip)
txtName_flip = txtPath.rsplit("\\",1)[-1].split(".txt")[0] + "_flipxy.txt"
txtPath_flip = os.path.join(txtPath.rsplit("\\",1)[0:-1][0], txtName_flip)
# 打开原来的txt标签文件，修改原来的x坐标为 xcenter’= 1 - xcenter
with open(txtPath, "r") as fsrc:
with open(txtPath_flip, "w") as f:
lines = fsrc.readlines()
for line in lines:
temp = line.split(" ")
label, xcenter, ycenter, objw, objh = temp[1], temp[3], temp[5], temp[7], temp[9]
if flag == 0:
xcenter = 1 - float(xcenter)
elif flag == 1:
ycenter = 1 - float(ycenter)
elif flag == 2:
xcenter = 1 - float(xcenter)
ycenter = 1 - float(ycenter)
f.write(" {} ".format(label))
f.write(" {} ".format(xcenter))
f.write(" {} ".format(ycenter))
f.write(" {} ".format(objw))
f.write(" {} ".format(objh))
f.write(" \n")
if flag == 0:
dst = cv2.flip(img, 1)
elif flag == 1:
dst = cv2.flip(img, 0)
elif flag == 2:
dst = cv2.flip(img, 1)
dst = cv2.flip(dst, 0)
cv2.imwrite(imagePath_flip, dst)
def imageAugment_gamma(self, imagePath, txtFolder, labeledImageFolder,gamma=2):
txtName = imagePath.rsplit("\\",1)[-1].rsplit(".",1)[0] + ".txt"
txtPath = os.path.join(txtFolder, txtName)
print("imageAugment_gamma: image path:", imagePath)
img = cv2.imread(imagePath)
imageName_gamma = imagePath.rsplit("\\",1)[-1].rsplit(".",1)[0] + "_gamma{}.".format(gamma) + imagePath.rsplit("\\",1)[-1].rsplit(".",1)[-1]
imagePath_gamma = os.path.join(labeledImageFolder, imageName_gamma)
txtName_gamma = txtPath.rsplit("\\",1)[-1].split(".txt")[0] + "_gamma{}.txt".format(gamma)
txtPath_gamma = os.path.join(txtPath.rsplit("\\",1)[0:-1][0], txtName_gamma)
shutil.copyfile(txtPath, txtPath_gamma)
print("source txtPath", txtPath)
print("dst txtPath", txtPath_gamma)
print("source imagePath", imagePath)
print("dst imagePath", imagePath_gamma)
table = []
for i in range(256):
table.append(((i / 255.0) ** gamma) * 255)
table = np.array(table).astype("uint8")
dst = cv2.LUT(img, table)
cv2.imwrite(imagePath_gamma, dst)
def devideTrainSetAndTestSet(self, imagePathList, saveFolder, prePath="", prop=0.8):
trainTxt = saveFolder + "/train.txt"
testTxt = saveFolder + "/test.txt"
with open(trainTxt, "w") as ftrain:
with open(testTxt, "w") as ftest:
for item in imagePathList:
imagePath = prePath + item.rsplit("\\",1)[-1]
if random.random() < prop:
ftrain.write(imagePath)
ftrain.write("\n")
else:
ftest.write(imagePath)
ftest.write("\n")
def getHelp(self):
print("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
print("Function:\t ")
print("parameter：")
print("\t _\t \t _")
print("return:")
print("\t _\t \t _")
print("================================================================================\n")
print("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
print("Function:\t getAllImagePath(folder,recursion=False)")
print("parameter：")
print("\t folder:\t \t search path")
print(
"\t recursion = False:\t \t default parameters,if recursion = True it will recursively search the folder")
print("return:")
print("\t imagePathLIst\t \t image file path list")
print("================================================================================\n")
print("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
print("Function:\t getAllJsonPath(self,folder)")
print("parameter：")
print("\t folder\t \t folde rcontains json file ")
print("return:")
print("\t jsonPathLIst\t \t json file path list")
print("================================================================================\n")
print("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
print("Function:\t fromJsonFileToSearchImage(jsonPathList,imageFolder,extendName = '.jpg')")
print("parameter：")
print("\t jsonPathList \t \t jsonPathList")
print("\t imageFolder\t \t 存放图片的文件夹")
print("\t extendName\t \t 图片扩展名")
print("return:")
print("\t imagePathList\t \t 标注过的图片的路径")
print("================================================================================\n")
print("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
print("Function:\t changeLabelFromJsonToTxt(self,jsonPath,txtSaveFolder)")
print("txt文件内的保存模式为：label xcenter ycenter width height,坐标全是相对坐标")
print("parameter：")
print("\t jsonPath:\t \t json file path")
print("\t txtSaveFolder\t \t txt file save folder")
print("return:")
print("\t None\t \t ")
print("================================================================================\n")
if __name__ == "__main__":
rootPath = r"K:\\imageData\\golden_pad\\bonding-balls-dataset\\"
srcImageFolder = rootPath + "image"# 存放原始图片的文件夹,有标注过的和没标注过的图片
jsonFolder = rootPath + "label"# labelme标注的json文件保存的文件夹
labeledImageFolder = rootPath + "/temp/labeledImage"# 筛选出所有标注过的图片
txtFolder = rootPath +"/temp/txtLabel"# txt文件保存的路径
objFolder = rootPath +"/temp/obj"# 把图片和对应的txt文件放在一起，可以直接拿去训练的那种
trainAndTestFileSaveFolder = rootPath +"/temp"
prePath = "bondingBall/obj/"
# 创建文件夹
try:
if not os.path.exists(labeledImageFolder):
os.makedirs(labeledImageFolder)
if not os.path.exists(txtFolder):
os.makedirs(txtFolder)
if not os.path.exists(objFolder):
os.makedirs(objFolder)
except:
print("os.makedirs failed!")
# 初始化类
tool = darknetTool()
# 获取所有原始标注文件
jsonPathList = tool.getAllJsonFilePath(jsonFolder)
# 从json标签文件复查标注
# for item in jsonPathList:
# tool.showLabelFromJson(item,imageFolder)
# 从json标注文件查找图片，并保存到labeledImageFolder
srcLabeledImagePathList = tool.fromJsonFileToSearchImage(jsonPathList, srcImageFolder)
tool.copyAllLabeledImageToNewFolder(srcLabeledImagePathList, labeledImageFolder)
# json格式转为txt格式
for item in jsonPathList:
tool.changeLabelFromJsonToTxt(item, txtFolder)
txtPathList = tool.getAllTxtPath(txtFolder)
# 获取在labeledImageFolder中的图片的路径
dstLabeledImagePathList = tool.getAllImagePath(labeledImageFolder)
# 从txt标签文件查看标注结果
# 所有标注的图片路径都在dstLabeledImagePathList，通过图片路径去找对应的txt文件
# 所有的txt文件都在txtFolder文件夹下
# for imagePath in dstLabeledImagePathList:
# tool.showLabelFromTxt(imagePath,txtFolder)
# 数据增强
# 均值滤波
for imagePath in dstLabeledImagePathList:
tool.imageAugment_smooth(imagePath, txtFolder, labeledImageFolder,n=3) # n=9是滤波核大小
# 水平翻转
for imagePath in dstLabeledImagePathList:
tool.imageAugment_flip(imagePath, txtFolder, labeledImageFolder, flag=0) # flag=0 水平翻转
# 竖直翻转
for imagePath in dstLabeledImagePathList:
tool.imageAugment_flip(imagePath, txtFolder, labeledImageFolder, flag=1) # flag=1 竖直翻转
# 中心对称
for imagePath in dstLabeledImagePathList:
tool.imageAugment_flip(imagePath, txtFolder, labeledImageFolder, flag=2) # flag=2 中心对称
# gamma变换
for imagePath in dstLabeledImagePathList:
#tool.imageAugment_gamma(imagePath, txtFolder, labeledImageFolder, gamma=0.5)
tool.imageAugment_gamma(imagePath, txtFolder, labeledImageFolder, gamma=0.8)
tool.imageAugment_gamma(imagePath, txtFolder, labeledImageFolder, gamma=1.2)
#tool.imageAugment_gamma(imagePath, txtFolder, labeledImageFolder, gamma=2)
#图片重命名
# 图片和标签重命名为00001.jpg 00001.txt.......（可选）
# n=5表示zfill（）的位数，n=5则为00001.jpg...，n=3则为001.jpg....
tool.renameAllLabeledImageAndTxt(txtFolder, labeledImageFolder, n=8, startN=0)
#检查转换的最终结果
dstLabeledImagePathList = tool.getAllImagePath(labeledImageFolder)
# for imagePath in dstLabeledImagePathList:
# tool.showLabelFromTxt(imagePath,txtFolder)
#划分训练集和测试集
imagePathList = tool.getAllImagePath(labeledImageFolder)
# prePath = "data/obj/"则train.txt文件中的路径就会是data/obj/00001.jpg,data/obj/00002.jpg......
# prop=0.8表示80%的数据用来训练，其余的做验证集
tool.devideTrainSetAndTestSet(imagePathList, trainAndTestFileSaveFolder, prePath=prePath, prop=0.9)
网友评论