火灾数据集处理

数据集

下载链接：https://github.com/steffensbola/furg-fire-dataset
内容：MP4的视频和对应的xml文件
问题：1.This XML was generated using OpenCv 2.4.9
2.xml文件中很多帧的图片没有坐标信息

目标：处理成VOC2007数据集格式

一，视频解帧

import cv2
import numpy as np
import os


def save_img():
    video_path = r"F:\Fire\furg_fire_dataset\mp4/"
    save_path = r"F:\Fire\furg_fire_dataset\img/"
    videos = os.listdir(video_path)
    print(len(videos),videos)
    for video_name in videos:
        file_name = video_name.split('.')[0]
        # print(file_name)
        folder_name = save_path + file_name
        # print(folder_name)
        # exit()
        os.makedirs(folder_name, exist_ok=True)
        print("="*10+'开始处理视频:%s'%video_name+"="*10)
        vc = cv2.VideoCapture(video_path + video_name)
        c = 0
        rval = vc.isOpened()

        while rval:

            rval, frame = vc.read()
            pic_path = folder_name + '/'
            if rval:
                cv2.imwrite(pic_path + str(c) + '.jpg', frame)
                cv2.waitKey(1)
                c += 1
            else:
                break
            print('='*10+'%s图片数量：%i'%(video_name,c)+'='*10)
        vc.release()
        print('save_success')
        print(folder_name)

save_img()

二，处理原始xml文件，使用库xml.etree.ElementTree和xml.dom.minidom解析我都没成功。发现里面的节点格式不对，我就换了种处理方式，把xml格式转成txt文本格式。

import os
import re
import re

path = r"F:\Fire\furg_fire_dataset\ann/"
save_path = r"F:\Fire\furg_fire_dataset\txt/"
xml_file = os.listdir(path)
for txt_file in xml_file:
    print('='*10+'开始处理：'+txt_file+'='*10)
    file = txt_file.split('.')[0]
    # print(file)
    new_file = file+'.txt'
    xml_dir = path+txt_file
    txt_dir = path+new_file
    print(xml_dir,txt_dir)
    # exit()
    os.rename(xml_dir,txt_dir)
    print('+'*10+'%s更改格式：'%txt_file+new_file+'+'*10)
    print(save_path+txt_file)
    # exit()
    pic_name = []
    box_data = {}
    index = ['0,','1','2','3','4','5','6','7','8','9']

    # with open(path+txt_file) as f:
    with open(txt_dir) as f:

        strs = f.readlines()
        # print(strs)
        count = 0
        for i in strs:
            # print(i)
            # print(i.strip())
            c = i.strip()[0]
            # print('===============',c)
            # print(type(c))
            t_f = i.strip()[-18:]
            im_file = i.strip()[-14:]
            # print(im_file)
            # print('t-f===================',t_f)
            # exit()
            if t_f == '</annotations></_>':
                box_data[count] = i.strip()
                count +=1
                # print('+++++++++++++++++++++++++++++++++++',i)
            elif im_file == '</frameNumber>':
                pic_name.append(i.strip())
                # print('===================================',i)
            else:
                print('出错%s'%i)

    # print("pic_name:",pic_name)
    # print("box_data:",box_data)

    dict = {}
    for i in range(len(pic_name)-1):
        pic = pic_name[i].split('</')
        # print(pic)
        # print(pic[0][16:])
        for j in range(len(box_data)):
            box = box_data[j].split('</_></a')
            # print(box)
            # print(box[0])
            if i == j:
                dict[pic[0][16:]] = box[0]
            # else:
            #     dict[pic[0][16:]] = 'null'


    print("pic_name:",pic_name)
    print("box_data:",box_data)
    print(dict)
    print(len(pic_name))
    print(len(box_data))


    # with open(save_path+txt_file,'w') as f:
    with open(save_path+new_file,'w') as f:
        for k in dict:
            f.write(str(k) + ' '+dict[k]+'\n')
    print('&'*15+'%s处理结束！'%txt_file+'&'*15)

三，处理视频解帧的图片，删除没有标签信息的图片

import os
import re
import glob


img_path = r"F:\Fire\furg_fire_dataset\img/"
txt_path = r"F:\Fire\furg_fire_dataset\txt/"

file_names = os.listdir(img_path)

for file_name in file_names:
    print('&'*10+'开始处理图片文件夹：'+file_name+'&'*10)

    # exit()
    img_file = []
    txt_file = []
    txt_dir = txt_path+file_name+'.txt'
    print("="*10+'处理txt文本：'+txt_dir+'='*10)

    with open(txt_dir) as f:
        strs = f.readlines()
        print('strs:',strs)
        for txt in strs:
            # print(txt)
            s = txt.strip().split(' ')
            # print(s,s[0],s[1])
            if s[1]=='</annotations></_>':
                img_file.append(s[0])
            else:
                txt_file.append(txt.strip())
        # print(txt_file)
        # print(len(txt_file),len(strs))
        print('错误图片:',img_file,len(img_file))
        with open(txt_dir,'w') as k:
            for t in txt_file:
                k.write(t+'\n')

    for im in os.listdir(os.path.join(img_path, file_name)):
        # print(im)
        # print(i)
        pic = im.split('.')[0]
        t_f = pic in img_file
        # print(t_f)

        if t_f:
            print('+'*12+'错误图片数：'+str(len(img_file))+'+'*12)
            print('+'*8+'删除图片：'+img_path +file_name+'/'+ im+'+'*8)
            os.remove(img_path+file_name+'/'+im)

四，把图片重命名并复制到空的文件夹

import os
import shutil


# 复制图片到新的文件夹
path = r"F:\Fire\furg_fire_dataset\img/"
xml_path = r"E:\xunleidownload\fire-dataset-dunnings\videos\furg-fire-dataset\Annotations/"
save_path = r"F:\Fire\furg_fire_dataset\pic/"

file_path = os.listdir(path)
print(len(file_path),file_path)

for i in file_path:
    img_files = os.listdir(path+i)
    print(i)
    for img in img_files:
        # print(img)
        src_dir = path+i+'/'+img
        dst_dir = path+i+'/'+i+img
        pic_dir = save_path+ i+img
        # print(img)
        print(src_dir)
        print(dst_dir)
        print(pic_dir)
        # exit()
        os.rename(src_dir,dst_dir)
        shutil.copyfile(dst_dir,pic_dir)

五，使用txt的标签信息，生成图片对应的xml文件

import os, sys
import glob
from PIL import Image

obj_name = 'fire'
src_img_dir = r"F:\Fire\furg_fire_dataset\img\HouseTwo"
src_txt_dir = r"F:\Fire\furg_fire_dataset\txt\HouseTwo.txt"
# src_xml_dir = r"F:\Fire\furg_fire_dataset\xml"
src_xml_dir = r"F:\Fire\furg_fire_dataset\xml"
# src_txt_dir = src_img_dir + 'F:\Fire\furg_fire_dataset\xml'
# src_xml_dir = src_img_dir

img_Lists = glob.glob(src_img_dir + '/*.jpg')
print(img_Lists)
img_basenames = []  # e.g. 100.jpg
for item in img_Lists:
    img_basenames.append(os.path.basename(item))

img_names = []  # e.g. 100
for item in img_basenames:
    temp1, temp2 = os.path.splitext(item)
    img_names.append(temp1)

for img in img_names:
    im = Image.open((src_img_dir + '/' + img + '.jpg'))
    print('img_name:',img)
    img_label = img.split('seTwo')[1]
    print('img_label:',img_label)
    # exit()
    width, height = im.size

    # open the crospronding txt file
    gt = open(src_txt_dir).read().splitlines()  # int x，int y，int width，int height  x,y是左上角
    print('boxes:',gt)
    # exit()
    # gt = open(src_txt_dir + '/gt_' + img + '.txt').read().splitlines()

    # write in xml file
    # os.mknod(src_xml_dir + '/' + img + '.xml')
    xml_file = open((src_xml_dir + '/' + img + '.xml'), 'w')
    xml_file.write('<annotation>\n')
    xml_file.write('    <folder>VOC2007</folder>\n')
    xml_file.write('    <filename>' + str(img) + '.jpg' + '</filename>\n')
    xml_file.write('    <size>\n')
    xml_file.write('        <width>' + str(width) + '</width>\n')
    xml_file.write('        <height>' + str(height) + '</height>\n')
    xml_file.write('        <depth>3</depth>\n')
    xml_file.write('    </size>\n')

    # write the region of image on xml file
    for img_each_label in gt:
        spt = img_each_label.split(' ')  # 这里如果txt里面是以逗号‘，’隔开的，那么就改为spt = img_each_label.split(',')。
        # print("========",spt)
        if spt[0] == img_label:
        # print(spt[0])
        # print(spt[-1])
        # print(int(float(int(spt[1])+int(spt[3]))))
        # print(int(spt[2])+int(spt[4]))
        # exit()
            xml_file.write('    <object>\n')
            # xml_file.write('        <name>' + str(spt[-5])+' '+str(spt[-6]) + '</name>\n')
            xml_file.write('        <name>' + obj_name + '</name>\n')
            xml_file.write('        <pose>Unspecified</pose>\n')
            xml_file.write('        <truncated>0</truncated>\n')
            xml_file.write('        <difficult>0</difficult>\n')
            xml_file.write('        <bndbox>\n')
            xml_file.write('            <xmin>' + str(int(float(spt[1]))) + '</xmin>\n')
            xml_file.write('            <ymin>' + str(int(float(spt[2]))) + '</ymin>\n')
            xml_file.write('            <xmax>' + str(int(float(int(spt[1])+int(spt[3])))) + '</xmax>\n')
            xml_file.write('            <ymax>' + str(int(float(int(spt[2])+int(spt[4])))) + '</ymax>\n')
            xml_file.write('        </bndbox>\n')
            xml_file.write('    </object>\n')

    xml_file.write('</annotation>')

六，VOC2007格式的JPEGImages和Annotations已备齐。还缺ImageSets\Main下的训练测试文本。

import os
import random
from glob import glob
import shutil

saved_path = r'F:\Fire\smoke\MP4\smoke_detect-PascalVOC-export\VOC2007/'
# saved_path = r"E:\xunleidownload\VOCdevkit\VOC2007/"
# trainval_percent = 0.66
trainval_percent = 0.8
# train_percent = 0.8
#
train_percent = 0.8
xmlfilepath = saved_path + 'Annotations/'
txtsavepath = saved_path + '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(saved_path + 'ImageSets/Main/trainval.txt', 'w')
ftest = open(saved_path + 'ImageSets/Main/test.txt', 'w')
ftrain = open(saved_path + 'ImageSets/Main/train.txt', 'w')
fval = open(saved_path + 'ImageSets/Main/val.txt', 'w')

for i  in list:
    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()