在只用VOC数据集训练得到的行人检测效果还是泛化效果一般，因此接下来尝试将多个数据集转成YOLO格式和VOC格式。

from pycocotools.coco import COCO
import shutil
import os
from lxml.etree import Element, SubElement, tostring
from xml.dom.minidom import parseString

# 获取所需要的类名和id
# path为类名和id的对应关系列表的地址（标注文件中可能有很多类，我们只加载该path指向文件中的类）
# 返回值是一个字典，键名是类名，键值是id
def get_classes_and_index(path):
    D = {}
    f = open(path)
    for line in f:
        temp = line.rstrip().split(',', 2)
        print("temp[0]:" + temp[0] + "\n")
        print("temp[1]:" + temp[1] + "\n")
        D[temp[1]] = temp[0]
    return D

def make_voc_dir():
    # labels 目录若不存在，创建labels目录。若存在，则清空目录
    if not os.path.exists('../VOC2007/Annotations'):
        os.makedirs('../VOC2007/Annotations')
    if not os.path.exists('../VOC2007/ImageSets'):
        os.makedirs('../VOC2007/ImageSets')
        os.makedirs('../VOC2007/ImageSets/Main')
    if not os.path.exists('../VOC2007/JPEGImages'):
        os.makedirs('../VOC2007/JPEGImages')

if __name__ == '__main__':
    make_voc_dir()
    VOCRoot = '../VOC2007'
    dataDir = '/home/common/wangsong/coco'  # COCO数据集所在的路径
    dataType = 'val2017'  # 要转换的COCO数据集的子集名
    annFile = '%s/annotations/instances_%s.json' % (dataDir, dataType)  # COCO数据集的标注文件路径
    classes = get_classes_and_index('/home/common/wangsong/coco/coco_list.txt') # {'person': '0'}

    coco = COCO(annFile)  # 加载解析标注文件

    imgIds = coco.getImgIds()  # 获取标注文件中所有图片的COCO Img ID
    catIds = coco.getCatIds()  # 获取标注文件总所有的物体类别的COCO Cat ID

    for imgId in imgIds:
        objCount = 0  # 一个标志位，用来判断该img是否包含我们需要的标注
        print('imgId :%s' % imgId)
        Img = coco.loadImgs(imgId)[0]  # 加载图片信息
        print('Img :%s' % Img)
        filename = Img['file_name']    # 获取图片名   '000000391895.jpg'
        width = Img['width']           # 获取图片尺寸
        height = Img['height']         # 获取图片尺寸 360
        print('filename :%s, width :%s ,height :%s' % (filename, width, height))
        annIds = coco.getAnnIds(imgIds=imgId, catIds=catIds, iscrowd=None)  # 获取该图片对应的所有COCO物体类别标注ID
        print('annIds :%s' % annIds)

        node_root = Element('annotation')
        node_folder = SubElement(node_root, 'folder')
        node_folder.text = 'JPEGImages'
        node_filename = SubElement(node_root, 'filename')
        node_filename.text = 'VOC2007/JPEGImages/%s' % filename
        node_size = SubElement(node_root, 'size')
        node_width = SubElement(node_size, 'width')
        node_width.text = '%s' % width
        node_height = SubElement(node_size, 'height')
        node_height.text = '%s' % height
        node_depth = SubElement(node_size, 'depth')
        node_depth.text = '3'
        
        for annId in annIds:
            anns = coco.loadAnns(annId)[0]  # 加载标注信息
            catId = anns['category_id']  # 获取该标注对应的物体类别的COCO Cat ID 1
            cat = coco.loadCats(catId)[0]['name']  # 获取该COCO Cat ID对应的物体种类名 'person'

            # 如果该类名在我们需要的物体种类列表中，将标注文件转换为YOLO需要的格式
            if cat in classes:
                objCount = objCount + 1

                cls_id = classes[cat] # 0
                box = anns['bbox']    # <class 'list'>: [339.88, 22.16, 153.88, 300.73]
                size = [width, height]
                xmin = int(box[0]) + 1
                ymin = int(box[1]) + 1
                obj_width = int(box[2])
                obj_height = int(box[3])
                xmax = xmin + obj_width
                ymax = ymin + obj_height

                difficult = 0
                if obj_height <= 4 or obj_width <= 4:
                    difficult = 1

                node_object = SubElement(node_root, 'object')
                node_name = SubElement(node_object, 'name')
                node_name.text = cat
                node_difficult = SubElement(node_object, 'difficult')
                node_difficult.text = '%s' % difficult
                node_bndbox = SubElement(node_object, 'bndbox')
                node_xmin = SubElement(node_bndbox, 'xmin')
                node_xmin.text = '%s' % xmin
                node_ymin = SubElement(node_bndbox, 'ymin')
                node_ymin.text = '%s' % ymin
                node_xmax = SubElement(node_bndbox, 'xmax')
                node_xmax.text = '%s' % xmax
                node_ymax = SubElement(node_bndbox, 'ymax')
                node_ymax.text = '%s' % ymax
                node_name = SubElement(node_object, 'pose')
                node_name.text = 'Unspecified'
                node_name = SubElement(node_object, 'truncated')
                node_name.text = '0'


        if objCount > 0:
            image_path = VOCRoot + '/JPEGImages/' + filename
            xml = tostring(node_root, pretty_print=True)  # 'annotation'
            dom = parseString(xml)
            xml_name = filename.replace('.jpg', '.xml')
            xml_path = VOCRoot + '/Annotations/' + xml_name
            with open(xml_path, 'wb') as f:
                f.write(xml)
            shutil.copy('%s/%s/%s' % (dataDir, dataType, filename), '../VOC2007/JPEGImages/' + filename)

这里需要注意的是

1. 要自己创建一个coco/coco_list.txt，里面内容如图所示

2. 需要运行两次，第一次训练集，第二次改为验证集

然后得到了下面的结果：