以遥感DIOR数据集为例，其标注文件为.xml格式，例

想把某一类从中取出来并生成针对此类的mask，实现方法是将.xml转化为json后读取object中的内容，将boundingbox的值取出生成mask图，需要用到的包如下

import simplejson
import xmltodict
import numpy as np

f.open()取出.xml数据，转换为json并读取字符

xmlparse = xmltodict.parse(xmlstr)
jsonstr = simplejson.dumps(xmlparse,indent=1)
simplejson_list = simplejson.loads(jsonstr, encoding='utf-8', strict=False)

取出object信息

annotation_objs = simplejson_list['annotation']['object']

由于DIOR数据集中一张图上可能有多个目标，故需要循环查找所需类别。在上个语句中，若一张图中只有一个目标，则annotation_objs为dict类型，若有多个目标则为list，所以要针对这两种情况分别讨论，不能直接循环查找。先判断是否为list，再循环list中每个obj的信息，将名字和boundingbox取出，对mask赋值

    if isinstance(annotation_objs, list): # 判断是否为List
        len_obj = len(annotation_objs)
        for i in range(len_obj):
            obj_name = annotation_objs[i]['name']
            if obj_name == obj_str:
                k = k+1
                obj_bnd = annotation_objs[i]['bndbox']
                xmin = int(obj_bnd['xmin'])
                ymin = int(obj_bnd['ymin'])
                xmax = int(obj_bnd['xmax'])
                ymax = int(obj_bnd['ymax'])

                mask[ymin:ymax, xmin:xmax] = 1

        
    else: # 若不为list则只有一个obj，为dict
        obj_name = annotation_objs['name']
        if obj_name == obj_str:
            k = k + 1
            obj_bnd = annotation_objs['bndbox']
            xmin = int(obj_bnd['xmin'])
            ymin = int(obj_bnd['ymin'])
            xmax = int(obj_bnd['xmax'])
            ymax = int(obj_bnd['ymax'])

            mask[ymin:ymax, xmin:xmax] = 1

其中k为是否含所需类别的flag，若k不等于0则说明本张图含有所需目标。
完整代码如下：

# -*- coding: utf-8 -*-
import simplejson
import xmltodict
import numpy as np
import cv2
import os
#定义xml转json的函数
def xmltojson(xmlstr, k, filename):
    xmlparse = xmltodict.parse(xmlstr)
    jsonstr = simplejson.dumps(xmlparse,indent=1)
    simplejson_list = simplejson.loads(jsonstr, encoding='utf-8', strict=False)

    mask_dir = './' # mask保存路径
    image_dir = './'  # 含所需目标的原图保存路径
    imgread_dir = './' # 原图读取路径
    mask = np.zeros((800, 800))

    annotation_objs = simplejson_list['annotation']['object']
    obj_str = 'ship' # 所需目标类别
    # 先判断图像中是否有多个目标
    if isinstance(annotation_objs, list):
        len_obj = len(annotation_objs)
        for i in range(len_obj):
            obj_name = annotation_objs[i]['name']
            if obj_name == obj_str:
                k = k+1
                obj_bnd = annotation_objs[i]['bndbox']
                xmin = int(obj_bnd['xmin'])
                ymin = int(obj_bnd['ymin'])
                xmax = int(obj_bnd['xmax'])
                ymax = int(obj_bnd['ymax'])

                mask[ymin:ymax, xmin:xmax] = 1


    else:
        obj_name = annotation_objs['name']
        if obj_name == obj_str:
            k = k + 1
            obj_bnd = annotation_objs['bndbox']
            xmin = int(obj_bnd['xmin'])
            ymin = int(obj_bnd['ymin'])
            xmax = int(obj_bnd['xmax'])
            ymax = int(obj_bnd['ymax'])

            mask[ymin:ymax, xmin:xmax] = 1

    if k != 0:
        name = os.path.splitext(filename)[0] + '.jpg'
        img_rgb = cv2.imread(imgread_dir +name)

        mask_name = mask_dir + name
        img_name = image_dir + name
        cv2.imwrite(mask_name, mask*255)
        cv2.imwrite(img_name, img_rgb)




if __name__=="__main__":
    for filename in os.listdir(r"./"):  # listdir的参数是.xml文件夹的路径
        print(filename)
        f = open('./' + filename) # 读取.xml
        data = f.read()
        k = 0
        xmltojson(data, k, filename)