#先改jpg名，再改xml名，类似DW001.jpg，DW002.xml。。
#
import os
class BatchRename():
'''''
批量重命名文件夹中的图片文件
'''
def __init__(self):
#我的图片文件夹路径horse
self.path = '/Users/sisyphus/TianChi/traindata/normalxml'
def rename(self):
filelist = os.listdir(self.path)
#filelist.sort()#+
filelist.sort(key=lambda x:int(x[-6:-4]))#以**.jpg排序
total_num = len(filelist)
i = 1
n = 6
for item in filelist:
# if item.endswith('.jpg'): ##################
if item.endswith('.xml'):
print(item)
n = 3 - len(str(i))
src = os.path.join(os.path.abspath(self.path), item)
# dst = os.path.join(os.path.abspath(self.path), str('normal')+str(0)*n + str(i) + '.jpg')############
dst = os.path.join(os.path.abspath(self.path), str('normal')+str(0)*n + str(i) + '.xml')
try:
os.rename(src, dst)
print('converting %s to %s ...' % (src, dst))
i = i + 1
except:
continue
print('total %d to rename & converted %d jpgs/xml' % (total_num, i-1))
if __name__ == '__main__':
demo = BatchRename()
demo.rename()
2. 修改xml内容：
from xml.etree import ElementTree
import os
jpgpath="/Users/sisyphus/TianChi/traindata/normal/"#末尾要加斜杠
jpglist = os.listdir(jpgpath)
jpglist.sort()
xmlpath="/Users/sisyphus/TianChi/traindata/normalxml/"########
xmllist = os.listdir(xmlpath)
xmllist.sort()
i=0
for item in xmllist:
# print(item)
xmldoc = ElementTree.parse(xmlpath+xmllist[i])
node = xmldoc.find('./filename')
node.text = jpglist[i]
node = xmldoc.find('object/name')
node.text = 'normal' #######################
##以下针对normal
node = xmldoc.find('object/bndbox/xmin')
node.text = '0'
node = xmldoc.find('object/bndbox/ymin')
node.text = '0'
node = xmldoc.find('object/bndbox/xmax')
node.text = '0'
node = xmldoc.find('object/bndbox/ymax')
node.text = '0'
##以上针对normal
xmldoc.write(xmlpath+xmllist[i])
i=i+1
print(item)
3 数据增强
import os
import cv2
import imghdr
import time
from PIL import Image, ImageEnhance, ImageOps, ImageFile
import numpy as np
class PreprocessIMG:
def __init__(self):
pass
#图像裁剪
@staticmethod
def img_resize(img_path):
im = cv2.imread(img_path)
img = cv2.resize(im, (416, 416))
return img
#图像增强（锐化）
@staticmethod
def img_resize_enhance(img_path):
im = cv2.imread(img_path)
# im = cv2.resize(im, (416, 416))
imgRGB = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)
PIL_Img = Image.fromarray(imgRGB)
PIL_ImgEn = ImageEnhance.Sharpness(PIL_Img).enhance(2.0)
return PIL_ImgEn
#图像增强（对比度）
@staticmethod
def img_resize_contrast(img_path):
im = cv2.imread(img_path)
# im = cv2.resize(im, (416, 416))
imgRGB = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)
PIL_Img = Image.fromarray(imgRGB)
contrast = ImageEnhance.Contrast(PIL_Img)#获取对比度对象
contrast_img = contrast.enhance(2.0)#增加对比度
return contrast_img
#图像降噪
@staticmethod
def img_resize_denoise(img_path):
im = cv2.imread(img_path)
# im = cv2.resize(im, (416, 416))
dst = cv2.fastNlMeansDenoisingColored(im,None,10,10,7,21)
return dst
#均值滤波
@staticmethod
def img_resize_blur(img_path):
im = cv2.imread(img_path)
# im = cv2.resize(im, (416, 416))
blur = cv2.blur(im,(3,5))#模板大小3*5
return blur
#高斯滤波
@staticmethod
def img_resize_gauss(img_path):
im = cv2.imread(img_path)
# im = cv2.resize(im, (416, 416))
blur = cv2.GaussianBlur(im,(5,5),0)
return blur
#中值滤波
@staticmethod
def img_resize_media(img_path):
im = cv2.imread(img_path)
# im = cv2.resize(im, (416, 416))
blur = cv2.medianBlur(im,5)
return blur
#双边滤波
@staticmethod
def img_resize_bil(img_path):
im = cv2.imread(img_path)
# im = cv2.resize(im, (416,416))
#滤波领域直径, 空间高斯函数标准差，灰度值相似性标准差
blur = cv2.bilateralFilter(im,9,75,75)
return blur
def ProcessIMG(path, newpath):
if os.path.isdir(path):
img_names = os.listdir(path)
else:
img_names = [path]
for img_name in img_names:
print(img_name)
tem_img_name = os.path.join(path, img_name)
if tem_img_name.split('.')[1] != "DS_Store":
# image = PreprocessIMG.img_resize(tem_img_name)
image = PreprocessIMG.img_resize_enhance(tem_img_name)
# image = PreprocessIMG.img_resize_contrast(tem_img_name)
# image = PreprocessIMG.img_resize_denoise(tem_img_name)
# image = PreprocessIMG.img_resize_blur(tem_img_name)
# image = PreprocessIMG.img_resize_gauss(tem_img_name)
# image = PreprocessIMG.img_resize_media(tem_img_name)
# image = PreprocessIMG.img_resize_bil(tem_img_name)
image = Image.fromarray(np.uint8(image))
image.save(newpath + '/' + str(img_name) ,format="jpeg")
if __name__ == '__main__':
# path = "/Users/sisyphus/py/testfrom1"
# newpath = "/Users/sisyphus/py/testto1"
path = "/Users/sisyphus/TianChi/traindata/JPEGraw"
# path = "/Users/sisyphus/SSD-Tensorflow/VOCtest2018/JPEGImages270Raw"
# newpath = "/Users/sisyphus/darkflow/VOCtest2018/JPEGImagesTrainAugEnh"
# newpath = "/Users/sisyphus/darkflow/VOCtest2018/JPEGImages270AugGau"
# newpath = "/Users/sisyphus/darkflow/VOCtest2018/JPEGImages270AugEnh"
# newpath = "/Users/sisyphus/SSD-Tensorflow/VOCtest2018/JPEGImages270Enh"#先建立文件夹
# newpath = "/Users/sisyphus/keras-yolo3/test2"
newpath = "/Users/sisyphus/TianChi/traindata/JPEGenh"
ProcessIMG(path, newpath)
4.修改增强集jpg名称
import re
import os
import time
#str.split(string)分割字符串
#'连接符'.join(list) 将列表组成字符串
def change_name(path):
global i
if not os.path.isdir(path) and not os.path.isfile(path):
return False
if os.path.isfile(path):
file_path = os.path.split(path) #分割出目录与文件
lists = file_path[1].split('.') #分割出文件与文件扩展名
file_ext = lists[-1] #取出后缀名(列表切片操作)
img_ext = ['bmp','jpeg','gif','psd','png','jpg']
if file_ext in img_ext:
os.rename(path,file_path[0]+'/'+lists[0]+'enh.'+file_ext)##########
i+=1 #注意这里的i是一个陷阱
#或者
#img_ext = 'bmp|jpeg|gif|psd|png|jpg'
#if file_ext in img_ext:
# print('ok---'+file_ext)
elif os.path.isdir(path):
for x in os.listdir(path):
print(x)
print(os.path.join(path,x))
change_name(os.path.join(path,x)) #os.path.join()在路径处理上很有用
#img_dir = "/Users/sisyphus/darkflow/VOCtest2018/JPEGImagesTrainAugEnh"
#img_dir = "/Users/sisyphus/darkflow/VOCtest2018/JPEGImagesTrainAugGau"
#img_dir = "/Users/sisyphus/darkflow/VOCtest2018/JPEGImages270AugCon"
#img_dir = "/Users/sisyphus/SSD-Tensorflow/VOCtest2018/JPEGImages270Enh"
#img_dir = '/Users/sisyphus/darkflow/VOCtest2018/testJPEG/'
#img_dir = img_dir.replace('\\','/')
img_dir = '/Users/sisyphus/TianChi/traindata/JPEGenh'
start = time.time()
i = 0
change_name(img_dir)
c = time.time() - start
print('程序运行耗时:%0.2f'%(c))
print('总共处理了 %s 张图片'%(i))
5修改增强集xml名称
import re
import os
import time
#str.split(string)分割字符串
#'连接符'.join(list) 将列表组成字符串
def change_name(path):
global i
if not os.path.isdir(path) and not os.path.isfile(path):
return False
if os.path.isfile(path):
file_path = os.path.split(path) #分割出目录与文件
lists = file_path[1].split('.') #分割出文件与文件扩展名
file_ext = lists[-1] #取出后缀名(列表切片操作)
img_ext = ['xml']
if file_ext in img_ext:
os.rename(path,file_path[0]+'/'+lists[0]+'enh.'+file_ext)#######++
i+=1 #注意这里的i是一个陷阱
#或者
#img_ext = 'bmp|jpeg|gif|psd|png|jpg'
#if file_ext in img_ext:
# print('ok---'+file_ext)
elif os.path.isdir(path):
for x in os.listdir(path):
print(x)
print(os.path.join(path,x))
change_name(os.path.join(path,x)) #os.path.join()在路径处理上很有用
#img_dir = "/Users/sisyphus/darkflow/VOCtest2018/AnnotationTrainAugEnh"
#img_dir = "/Users/sisyphus/darkflow/VOCtest2018/Annotations270AugCon"#####先修改上面++
#img_dir = "/Users/sisyphus/SSD-Tensorflow/VOCtest2018/Annotations270Enh"
#img_dir = '/Users/sisyphus/darkflow/VOCtest2018/testAnn/'
#img_dir = img_dir.replace('\\','/')
img_dir = '/Users/sisyphus/TianChi/traindata/xmlenh'
start = time.time()
i = 0
change_name(img_dir)
c = time.time() - start
print('程序运行耗时:%0.2f'%(c))
print('总共处理了 %s 份xml'%(i))
6 修改增强集xml内容
from xml.etree import ElementTree
import os
jpgpath="/Users/sisyphus/TianChi/traindata/JPEGenh/"#末尾要加斜杠
jpglist = os.listdir(jpgpath)
jpglist.sort()
xmlpath="/Users/sisyphus/TianChi/traindata/xmlenh/"########
xmllist = os.listdir(xmlpath)
xmllist.sort()
i=0
for item in xmllist:
# print(item)
xmldoc = ElementTree.parse(xmlpath+xmllist[i])
node = xmldoc.find('./filename')
node.text = jpglist[i]
# node = xmldoc.find('object/name')
# node.text = 'normal' #######################
# ##以下针对normal
# node = xmldoc.find('object/bndbox/xmin')
# node.text = '0'
# node = xmldoc.find('object/bndbox/ymin')
# node.text = '0'
# node = xmldoc.find('object/bndbox/xmax')
# node.text = '0'
# node = xmldoc.find('object/bndbox/ymax')
# node.text = '0'
# ##以上针对normal
xmldoc.write(xmlpath+xmllist[i])
i=i+1
print(item)
7 修改Main/txt文件夹
classes='DW'
classes='MB'
classes='ZD'
classes='normal'
#xmlfilepath = '/Users/sisyphus/keras-retinanet/VOCdevkit/VOC2018/Annotations/'
#txtsavepath = '/Users/sisyphus/keras-retinanet/VOCdevkit/VOC2018/ImageSets/Main/'
xmlfilepath = '/Users/sisyphus/TianChi/VOC2018/Annotations/'
txtsavepath = '/Users/sisyphus/TianChi/VOC2018/ImageSets/Main/'
total_xml = os.listdir(xmlfilepath)
num=len(total_xml)
list=range(num)
tv=int(num*trainval_percent)
tr=int(tv*train_percent)
trainval= random.sample(list,tv)
train=random.sample(trainval,tr)
ftrainval = open(txtsavepath+'_trainval.txt', 'w')
ftest = open(txtsavepath+'_test.txt', 'w')
ftrain = open(txtsavepath+'_train.txt', 'w')
fval = open(txtsavepath+'_val.txt', 'w')
for i in list:
# name='/Users/sisyphus/keras-retinanet/VOCdevkit/VOC2018/JPEGImages/'+total_xml[i][:-4]+'\n'
name=total_xml[i][:-4]+'\n'
if i in trainval:
ftrainval.write(name)
if i in train:
ftrain.write(name)
else:
fval.write(name)
else:
ftest.write(name)
ftrainval.close()
ftrain.close()
fval.close()
ftest .close()
8 修改names文件
在YOLO主目录的data文件夹下，创建一个.names文件，文件名任意。比如mydata.names。在该文件中写入所有类别的名称，每一类占一行。
9 修改data文件
接下来我们要做的是修改YOLO的cfg文件。在darknet的主目录下，进入cfg目录，找到voc.data打开，修改其中的内容
classes= 1 #训练数据的类别数目，我这里只有一类，所以这里是1
train = <path-to-voc>/train.txt #上面1.2步骤生成的train文件路径
valid = <path-to-voc>2007_test.txt #上面1.2步骤生成的val文件路径
names = data/voc.names #上面2.1步骤创建的names文件路径
backup = backup #这是训练得到的model的存放目录，建议自己修改。
train = <path-to-voc>/train.txt #上面1.2步骤生成的train文件路径
valid = <path-to-voc>2007_test.txt #上面1.2步骤生成的val文件路径
names = data/voc.names #上面2.1步骤创建的names文件路径
backup = backup #这是训练得到的model的存放目录，建议自己修改。
10 修改cfg
把cfg文夹中的yolov3-voc.cfg复制到自己项目目录下，并修改：
[net]
# Testing
# batch=1
# subdivisions=1 #训练时候把上面Testing的参数注释
# Training
batch=64
subdivisions=32 #这个参数根据自己GPU的显存进行修改，显存不够就改大一些
... #因为训练时每批的数量 = batch/subdivisions
...
...
learning_rate=0.001 #根据自己的需求还有训练速度学习率可以调整一下
burn_in=1000
max_batches = 30000 #根据自己的需求还有训练速度max_batches可以调整一下
policy=steps
steps=10000,20000 #跟着max_batches做相应调整
...
...
...
[convolutional]
size=1
stride=1
pad=1
filters=30 #filters = 3*(classes + 5)
activation=linear
[yolo]
mask = 0,1,2
anchors = 10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326
classes=5 #修改类别数
num=9
jitter=.3
ignore_thresh = .5
truth_thresh = 1
random=1 #显存小的话 =0
#这个文件的最下面有3个YOLO层，这里我才放上来了一个，这三个地方的classes做相应修改
#每个YOLO层的上一层的convolutional层的filters也要修改wget https://pjreddie.com/media/files/darknet53.conv.7412 训练
./darknet detector train cfg/voc.data cfg/yolov3-voc.cfg darknet53.conv.74报错：
Cannot load image /home/ubuntu/fddb/JPEGImages/2002_08_07_big_img_1348.jpg STB Reason: can't fopen
查得可能需要opencv=1，
opencv=1 makefile 报错
fatal error: opencv2/highgui/highgui_c.h: No such file or directory
还在查找原因中。。
