在跑通Faster-rcnn的demo,以及机智地在VOC2007上只使用行人对象训练跑通后，还是要自己制作数据集。以Caltech行人数据集为例，要将下载的seq转为png格式，vbb格式转为xml格式，转变为VOC格式，还要读取pascal_voc.py中数据接口，保证数据读取正确。
代码主要参考这篇博客

1.seq转为png(python3)

import os
import glob
import cv2 as cv
 
# 保存图片以指定名字到指定位置
def save_img(dname, fn, i, frame):
    cv.imwrite('{}/{}_{}_{}.png'.format(
        out_dir, os.path.basename(dname),
        os.path.basename(fn).split('.')[0], i), frame)
    
    
#输出图片位置    
out_dir = 'E:/images'
if not os.path.exists(out_dir):
    os.makedirs(out_dir)
#seq文件位置
for dname in sorted(glob.glob('E:/bishe/database/set*')):
    for fn in sorted(glob.glob('{}/*.seq'.format(dname))):
        cap = cv.VideoCapture(fn)
        i = 0
        while True:
            # ret为标志位，bool型，是否读到数据，frame为视频帧图像数据
            ret, frame = cap.read()
            if not ret:
                break
            save_img(dname, fn, i, frame)
            i += 1
        print(fn)

注意：输出路径不能包括中文，python3中cv包好像读不了。

2.vbb转为txt(matlab带code3.2.1的库)

a = [0 1 2 3 4 5];  
vPath = 'E:\bishe\database\datasize caltech\annotations\';  
outpath = 'E:\bishe\database\datasize caltech\annotations\caltechAnnotations';   
if ~isdir(outpath)  
    mkdir(outpath);  
end  
disp('Begin to extract annotations from vbb....')  
for ii = 1:length(a)  
    vName = ['set0',num2str(a(ii))];    
    vbbpath = [vPath vName];  
    str = dir(vbbpath);    %读文件
    for k = 3:numel(str)  
        finalpath = [vbbpath,'\',str(k).name];   
        disp(finalpath);  
        A = vbb( 'vbbLoad', finalpath);    
        Path = [outpath,'\',vName];  
        if ~isdir(Path)  
            mkdir(Path);  
        end  
        st = str(k).name;  
        p = findstr(st,'.');  
        st = st(1,1:p(1)-1);  
        fnm = [st,'.txt'];       
        c=fopen([Path,'\',fnm],'w');    
        for i = 1:A.nFrame    
            iframe = A.objLists(1,i);    
            iframe_data = iframe{1,1};    
            n1length = length(iframe_data);    
            for  j = 1:n1length    
                iframe_dataj = iframe_data(j);    
                if iframe_dataj.pos(1) ~= 0  %pos  posv    
                    fprintf(c,'%d %f %f %f %f\n', i, iframe_dataj.pos(1),...  
                     iframe_dataj.pos(2),iframe_dataj.pos(3),iframe_dataj.pos(4));    
                end    
            end    
        end    
    end  
fclose(c);   
end  
disp('Done................')  

A便是vbb文件的信息，需要把断点打在A = vbb( 'vbbLoad', finalpath);

3.txt转为xml

这一段处理最费时间，基于原博客的代码我进行了很多修改。

from lxml.etree import Element, SubElement, tostring
import pprint
from xml.dom.minidom import parseString
import os
import pdb 
 
def mkdir(path):
    import os
 
    path = path.strip()
    path = path.rstrip("\\")
 
    isExists = os.path.exists(path)
    if not isExists:
        os.makedirs(path)
        print (path + 'ok')
        return True
    else:
        print (path + 'failed!')
        return False
 
def generate_xml(file_info, obj):
    # xml写操作
    node_root = Element('annotation')
 
    node_folder = SubElement(node_root, 'folder')
    node_folder.text = file_info[0]
 
    node_filename = SubElement(node_root, 'filename')
    node_filename.text = file_info[1]
 
    node_size = SubElement(node_root, 'size')
    node_width = SubElement(node_size, 'width')
    node_width.text = '640'
 
    node_height = SubElement(node_size, 'height')
    node_height.text = '480'
 
    node_depth = SubElement(node_size, 'depth')
    node_depth.text = '3'
 
    for obj_i in obj:
        print (obj_i)
        node_object = SubElement(node_root, 'object')
        node_name = SubElement(node_object, 'name')
        #node_name.text = 'mouse'
        node_name.text = 'person'
 
        node_bndbox = SubElement(node_object, 'bndbox')
        node_xmin = SubElement(node_bndbox, 'xmin')
        #node_xmin.text = '99'
        node_xmin.text = obj_i['xmin']
 
        node_ymin = SubElement(node_bndbox, 'ymin')
        #node_ymin.text = '358'
        node_ymin.text = obj_i['ymin']
 
        node_xmax = SubElement(node_bndbox, 'xmax')
        #node_xmax.text = '135'
        node_xmax.text = obj_i['xmax']
 
        node_ymax = SubElement(node_bndbox, 'ymax')
        #node_ymax.text = '375'
        node_ymax.text = obj_i['ymax']
 
    xml = tostring(node_root, pretty_print=True)  #格式化显示，该换行的换行
    dom = parseString(xml)
    file_root = 'E:/bishe/database/CaltechVOC/Annotations'
 
    fw = open(file_root+"/"+file_info[1].split('.')[0]+".xml", 'wb')
 
    fw.write(xml)
    print ("xml _ ok")
    fw.close()
 
    #for debug
    #print xml
 
def printPath(level, path):
    global allFileNum
    ''''' 
    打印一个目录下的所有文件夹和文件 
    '''
    # 所有文件夹，第一个字段是次目录的级别
    dirList = []
    # 所有文件
    fileList = []
    # 返回一个列表，其中包含在目录条目的名称(google翻译)
    files = os.listdir(path)
 #   pdb.set_trace() 
    # 先添加目录级别
    dirList.append(str(level))
    
    #获取文件，set00-set0* 
    for f in files:
        if(os.path.isdir(path + '/' + f)):
            # 排除隐藏文件夹。因为隐藏文件夹过多
            if(f[0] == '.'):
                pass
            else:
                # 添加非隐藏文件夹
                dirList.append(f)
        if(os.path.isfile(path + '/' + f)):
            # 添加文件
            fileList.append(f)
         
    # 当一个标志使用，文件夹列表第一个级别不打印
    i_dl = 0
    
    #获取每个文件下的txt文件
    for dl in dirList:
        if(i_dl == 0):
            i_dl = i_dl + 1
        else:
            # 打印至控制台，不是第一个的目录
            path1 = path + '/' + dl
            # 遍历获取文件下所有的txt文件
            files = os.listdir(path1)
            # 读取txt文件
            for fl in files:
                #if(os.path.isfile(path1 + '/' + f)):
                fileList.append(fl)
                file_info = []
                print(path1)
                file_name = path1+"/"+fl
                fw = open(file_name, 'r');
                #读取文件每一行信息
                line_content = fw.readlines()
                fw.close()
                print(line_content)
          
         
                tmp = -1
                obj = []
                con_len = len(line_content)
                try:
                    # 分割，每一行相当于一个单元
                    string = line_content[0].split(" ")
                    tmp = int(string[0])
                except Exception:
                    continue
                # folder的命名
                file_info.append('Caltech_{}_{}'.format(
                os.path.basename(dl),os.path.basename(fl).split('.')[0]))
                # 图片的命名需要与图片名对应
                file_info.append('{}_{}_{}.png'.format(
                        os.path.basename(dl),os.path.basename(fl).split('.')[0], str(tmp-1)))
                #box赋值，xmin,ymin不能等于0，否则faster-rcnn中数据溢出
                if int(float(string[1]))==0:
                    xmin = '1'
                else:
                    xmin = str(int(float(string[1])))
                if int(float(string[2]))==0:
                    ymin = '1'
                else:
                    ymin = str(int(float(string[2])))
                xmax = str(int(float(string[1]) + float(string[3])))
                ymax = str(int(float(string[2]) + float(string[4])))
                dict1 = {}
                dict1["xmin"] = xmin
                dict1["ymin"] = ymin
                dict1["xmax"] = xmax
                dict1["ymax"] = ymax
                obj.append(dict1)
                for con_i in range(1, con_len):
                    string = line_content[con_i].split(" ")
                    tmp1 = int(string[0])
                    #一帧中多个对象
                    if tmp == tmp1:
                        if int(float(string[1]))==0:
                            xmin = '1'
                        else:
                            xmin = str(int(float(string[1])))
                        if int(float(string[2]))==0:
                            ymin = '1'
                        else:
                            ymin = str(int(float(string[2])))
                        xmax = str(int(float(string[1]) + float(string[3])))
                        ymax = str(int(float(string[2]) + float(string[4])))
                        dict1 = {}
                        dict1["xmin"] = xmin
                        dict1["ymin"] = ymin
                        dict1["xmax"] = xmax
                        dict1["ymax"] = ymax
                        obj.append(dict1)
                        # 最后1帧不会再执行tmp1 > 0产生xml,因此需特殊处理
                        if con_i == con_len-1:
                            generate_xml(file_info, obj)
                    elif tmp1 > 0:             
                        generate_xml(file_info, obj)
                        #下一帧处理
                        obj = []
                        tmp = tmp1
                        file_info[1] = "{}_{}_{}.png".format(
                        os.path.basename(dl),os.path.basename(fl).split('.')[0], str(tmp-1))
                        if int(float(string[1]))==0:
                            xmin = '1'
                        else:
                            xmin = str(int(float(string[1])))
                        if int(float(string[2]))==0:
                            ymin = '1'
                        else:
                            ymin = str(int(float(string[2])))
                        xmax = str(int(float(string[1]) + float(string[3])))
                        ymax = str(int(float(string[2]) + float(string[4])))
                        dict1 = {}
                        dict1["xmin"] = xmin
                        dict1["ymin"] = ymin
                        dict1["xmax"] = xmax
                        dict1["ymax"] = ymax
                        obj.append(dict1)
                continue
 
def read_annotations_generate_fileinfo_obj(file_path):
    pass
 
if __name__=="__main__":
 
    #
    # file_info = ['set00/V000', '1.jpg']
    #
    # obj = []
    # obj1 = {"xmin":"1", "ymin":"1", "xmax":"5", "ymax":"5"}
    # obj2 = {"xmin":"2", "ymin":"2", "xmax":"6", "ymax":"6"}
    # obj.append(obj1)
    # obj.append(obj2)
    #
    # generate_xml(file_info, obj)
    #
 
    printPath(1, "E:/bishe/database/datasize caltech/annotations/caltechAnnotations")

注意点：

• matlab处理从1开始，而图片命名从0开始,因此读取txt中帧数1，需要做减1处理。

• box赋值，xmin,ymin不能等于0，因为faster-rcnn有-1处理，否则faster-rcnn中数据溢出。

  
  
  
  

4.产生测试集训练集

import os
import glob
import cv2 as cv
import random
import pdb 
 
 
def save_txt(nameList):
    
    trainval_percent = 0.66
    train_percent = 0.5
    folder_root = 'E:/bishe/database/CaltechVOC/'
    # pdb.set_trace() 
    num = len(nameList)
 
    #print num
    # 通过随机函数产生训练集测试集
    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(folder_root + 'ImageSets/Main/trainval.txt', 'a')
    ftest     = open(folder_root + 'ImageSets/Main/test.txt'    , 'a')
    ftrain    = open(folder_root + 'ImageSets/Main/train.txt'   , 'a')
    fval      = open(folder_root + 'ImageSets/Main/val.txt'     , 'a')
 
    for i in list:
        name = nameList[i] + '\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()
 
    
out_dir = 'E:/images'
nameList=[]
if not os.path.exists(out_dir):
    os.makedirs(out_dir)
for dname in sorted(glob.glob('E:/bishe/database/set*')):
    for fn in sorted(glob.glob('{}/*.seq'.format(dname))):
        cap = cv.VideoCapture(fn)
        i = 0
        while True:
            ret, frame = cap.read()
            if not ret:
                break
            name='{}_{}_{}'.format(
                    os.path.basename(dname),os.path.basename(fn).split('.')[0], i)
            # 遍历seq文件后获取所有图片名字
            nameList.append(name)      
            i += 1
save_txt(nameList)