總結訓練SSD時的小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檔案名：

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可視化：

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)

根據全部資料集Annotations随機生成訓練資料集和測試資料集的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')

總結訓練SSD時的小trick

繼續閱讀

簡單文檔分類——樸素貝葉斯算法樸素貝葉斯算法簡單文檔分類執行個體步驟總結樸素貝葉斯分類調用(sklearn)

考證大全 | 證券從業資格考試

敲黑闆！2021年證券從業考試考點預測

2021年銀行從業考試考情介紹,果斷收藏!

證券從業合格證書什麼時候列印？有哪些注意事項？

【幹貨滿滿】初級銀行從業考試《個人理财》重點梳理

2020年經濟師考試，難嗎？

初級銀行從業資格證有什麼用？

MBA提前面試純幹貨分享

MBA值得學麼

吳恩達logistic回歸實作

【人工智能行業大師訪談1】吳恩達采訪 Geoffery Hinton

深度學習模型分析人類複雜疾病的準确性

【趨高機器視覺】機器視覺技術原了解析及解決方案

解碼器用于語義分割：資料依賴的解碼可以實作靈活的特征聚合

cs231n斯坦福基于卷積神經網絡的CV學習筆記（一）KNN和線性分類器/分類器損失/反向傳播一，KNN圖像分類算法二，線性分類器三，線性分類器損失四，反向傳播五，神經網絡