总结训练SSD时的小trick

删除xml中某些obj：

import os  
import os.path  
import xml.dom.minidom  
from xml.dom.minidom import parse
import xml.dom.minidom
import os,shutil
import numpy as np
import cv2
from PIL import Image, ImageDraw

path="/data_1/SSD/cpls/cpls/xml_ls"  

files=os.listdir(path)
s=[]  
k=0
c=0
n=0
s=" " 
for xmlFile in files:
    k=k+1
    c=0
    print k
    print xmlFile
    if not os.path.isdir(xmlFile): 
        #print xmlFile  
        dom=xml.dom.minidom.parse(os.path.join(path,xmlFile))
        root=dom.documentElement  
        objects = root.getElementsByTagName("object")
        for object_ in objects:
            #a=object_.getElementsByTagName("name")[0].childNodes[0].nodeValue
            #if a=="fadongjifanghuzhuangzhi":
             #   print xmlFile  
             #   name=object_.getElementsByTagName('name')
              #  n0=name[0]
               # n0.firstChild.data='xcjly'
            xmin = object_.getElementsByTagName("xmin")[0].childNodes[0].nodeValue
            xmax = object_.getElementsByTagName("xmax")[0].childNodes[0].nodeValue
            ymin = object_.getElementsByTagName("ymin")[0].childNodes[0].nodeValue
            ymax = object_.getElementsByTagName("ymax")[0].childNodes[0].nodeValue
            #print (int(xmax)-int(xmin))
            #c=c+1
        #if(c>1 or c==0):
         #   s=s+xmlFile+" "
          #  n=n+1
            if (int(xmax)-int(xmin))>100 or (int(ymax)-int(ymin))>100:
                object_.parentNode.removeChild(object_)
               
            with open(os.path.join(path, xmlFile), 'w') as fh:
                dom.writexml(fh)
#print n
#print s

修改xml和img文件名：

import os
import shutil

## datadir AND savedir CAN NOT BE SAME
img_datadir="old/JPEGImages/"
img_savedir="new/JPEGImages/"

xml_datadir="old/Annotations/"
xml_savedir="new/Annotations/"

imglist=os.listdir(img_datadir)
k=0
for img in imglist:
 name=img
 k=k+1
 print k
# zzname=img.split("_",2)
 img_datapath=img_datadir+name
 xml_datapath=xml_datadir+name.replace("jpg","xml")
# img_save_name='mhq_171215_'+str(k)+"_"+zzname[2]
# xml_save_name=img_save_name.replace(".jpg",".xml")

 img_save_name='mhq_181105_'+str(k)+".jpg"
 xml_save_name=img_save_name.replace(".jpg",".xml")

 img_savepath=img_savedir+img_save_name 
 xml_savepath=xml_savedir+xml_save_name 

 #if os.path.exists(img_datapath):
 shutil.copy(img_datapath, img_savepath)
  
 #if os.path.exists(xml_datapath):
 shutil.copy(xml_datapath, xml_savepath)  

修改xml中filename：

# -*- coding:utf-8 -*-
import os  
import os.path  
import xml.dom.minidom  
from xml.dom.minidom import parse
import xml.dom.minidom
import os,shutil
#import numpy as np
#import cv2
import urllib
from PIL import Image, ImageDraw

path="new/Annotations/"  
files=os.listdir(path)
s=[]
num=0
for xmlFile in files:
    num+=1
    print(num)
    imgname=xmlFile.replace(".xml",".jpg")
    if not os.path.isdir(xmlFile): 
        print (xmlFile)
        dom=xml.dom.minidom.parse(os.path.join(path,xmlFile))
        root=dom.documentElement  
        #filename = root.getElementsByTagName("filename").childNodes[0].nodeValue
        #print (filename)
        filename1=root.getElementsByTagName('filename')
        n0=filename1[0]
        print (n0.firstChild.data)

	
        a=imgname
        n0.firstChild.data=a
        print (n0.firstChild.data)
    

        with open(os.path.join(path, xmlFile), 'w') as fh:
            dom.writexml(fh)

修改xml中objname：

import os  
import os.path  
import xml.dom.minidom  
from xml.dom.minidom import parse
import xml.dom.minidom
import os,shutil
import numpy as np
import cv2
from PIL import Image, ImageDraw

path="/xml"  

files=os.listdir(path)
s=[]  
k=0
for xmlFile in files:
    k=k+1
    print k
    print xmlFile
    if not os.path.isdir(xmlFile): 
        #print xmlFile  
        dom=xml.dom.minidom.parse(os.path.join(path,xmlFile))
        root=dom.documentElement  
        objects = root.getElementsByTagName("object")
        for object_ in objects:
            #a=object_.getElementsByTagName("name")[0].childNodes[0].nodeValue
            #if a=="fdjh":
            #    print xmlFile  
            name=object_.getElementsByTagName('name')
            n0=name[0]
            n0.firstChild.data='cpls'
               
            with open(os.path.join(path, xmlFile), 'w') as fh:
                dom.writexml(fh)

xml可视化：

from xml.dom.minidom import parse
import matplotlib.pyplot as plt
import xml.dom.minidom
import os,shutil
import matplotlib  
import numpy as np
import cv2
from PIL import Image, ImageDraw
##########################################################
root="/data_1/SSD/caffe/data/VOCdevkit/mydataset/"
#only need to change these
##########################################################
#annroot=root+'2/'
#picroot=root+'1/'
#annroot=root+'xml/'
#picroot=root+'img/'
annroot=root+'Annotations/'
picroot=root+'JPEGImages/'
anns=os.listdir(annroot)
imgs=os.listdir(picroot)

labelmap=["cpls"]

colormap=["red" , "green", "blue" , "yellow", "pink" , "olive" , "deeppink" , "darkorange", "purple", "cyan","red" , "green", "blue" , "yellow", "pink" , "olive" , "deeppink" , "darkorange", "purple", "cyan","red"]

def mkdir(path): 
	folder = os.path.exists(path)
	if not folder:
		os.makedirs(path)

number = 0
nn=0
for ann in anns:
    number += 1
    print (number)
    print (ann)
    annpath=annroot+ann
    picpath=picroot+ann.replace("xml","jpg")
    im = Image.open(picpath)
    img = cv2.imread(picpath)
    draw = ImageDraw.Draw(im)
    DOMTree = xml.dom.minidom.parse(annpath)
    collection = DOMTree.documentElement
    objects = collection.getElementsByTagName("object")
    labelsss = ""
    for object_ in objects:
        #print (object_)
        a=object_.getElementsByTagName("name")[0].childNodes[0].nodeValue
        k=a.split('.',1)
        kk=k[0]        
        b=str(kk) 	
        for i in range(0,len(labelmap)):
            label = labelmap[i]
            print (label)
	    
            if b == label:
	        nn+= 1
		if label not in labelsss:
    		    labelsss+= label +"_"
                bndboxs = object_.getElementsByTagName("bndbox")
                for bndbox in bndboxs:
                    xmin = bndbox.getElementsByTagName('xmin')[0].childNodes[0].nodeValue
                    ymin = bndbox.getElementsByTagName('ymin')[0].childNodes[0].nodeValue
                    xmax = bndbox.getElementsByTagName('xmax')[0].childNodes[0].nodeValue
                    ymax = bndbox.getElementsByTagName('ymax')[0].childNodes[0].nodeValue
                    xtmp1=xmin.split('.',1)
                    xmin1=xtmp1[0]
                    xtmp2= xmax.split('.', 1)
                    xmax1 = xtmp2[0]
                    xtmp3=ymin.split('.',1)
                    ymin1=xtmp3[0]
                    xtmp4=ymax.split('.',1)
                    ymax1=xtmp4[0]
                xmin = int(xmin1)
                ymin = int(ymin1)
                xmax = int(xmax1)
                ymax = int(ymax1)
                if xmin<0:
                    xmin=0
                if ymin<0:
                    ymin=0
                sp=img.shape
                if xmax>sp[1]:
                    xmax=sp[1]
                if ymax>sp[0]:
                    ymax=sp[0]


		roiimg=img[ymin: ymax, xmin:xmax]
		save_op = root+'chcc/'+label +"_"+"/"
		mkdir(save_op)
		saveopath = save_op+str(nn)+"_"+ann.replace("xml","jpg")		
		cv2.imwrite(saveopath,roiimg)

                draw.rectangle((xmin, ymin, xmax, ymax), outline = colormap[i])
                draw.rectangle((xmin-1, ymin-1, xmax-1, ymax-1), outline = colormap[i])
            	draw.rectangle((xmin+1, ymin+1, xmax+1, ymax+1), outline = colormap[i])
            	draw.rectangle((xmin-2, ymin-2, xmax-2, ymax-2), outline = colormap[i])
            	draw.rectangle((xmin+2, ymin+2, xmax+2, ymax+2), outline = colormap[i])
            	draw.rectangle((xmin-3, ymin-3, xmax-3, ymax-3), outline = colormap[i])
            	draw.rectangle((xmin+3, ymin+3, xmax+3, ymax+3), outline = colormap[i])	
            	break


        label_has=0
        for label in labelmap:
            if b != label:
                label_has = 1
        if not label_has:
            print (ann+"======"+b+"============================")
       

    
    save_p = root+'check/'+labelsss+"/"
    savepath = save_p+ann.replace("xml","jpg")
    mkdir(save_p)
    im.save(savepath)
        #cv2.imwrite(savepath,roiimg)

根据全部数据集Annotations随机生成训练数据集和测试数据集的txt：

make_txt.sh

cd /Annotations

find ./ -name "*" |sort >>../ImageSets/Main/all.txt

cd ../ImageSets/Main

python2 make_txt.py

make_txt.py

import os

num = 0
for tmpline in open("./all.txt"):
    num += 1
#    print num
    if num==1:
        continue
    tmp = tmpline.strip('\n')
    tmp = tmp.replace(".xml","")
    tmp = tmp.replace("./","")
    tmpp=tmp.split('_',3)
    number= int(tmpp[2])
    aa = number%10
    if aa==0 or aa==1:
	with open('/data_2/data/train_data/ss_big_obj_object_bdf/ssd_big_obj/ImageSets/Main/test.txt', 'a') as f:
	    f.writelines(tmp + '\n')
    elif aa==2:
	with open('/data_2/data/train_data/ss_big_obj_object_bdf/ssd_big_obj/ImageSets/Main/val.txt', 'a') as f:
	    f.writelines(tmp + '\n')
    else:
	with open('/data_2/data/train_data/ss_big_obj_object_bdf/ssd_big_obj/ImageSets/Main/trainval.txt', 'a') as f:
	    f.writelines(tmp + '\n')