数据处理：csv标注转换成xml

test_0.jpg,1653,1290,1773,1535,object,2448,3264
test_0.jpg,1485,1221,1648,1544,object,2448,3264
test_0.jpg,1345,1295,1481,1540,object,2448,3264
test_0.jpg,1221,1290,1341,1543,object,2448,3264
test_0.jpg,1079,1332,1216,1537,object,2448,3264
test_0.jpg,927,1285,1069,1531,object,2448,3264
test_0.jpg,679,1279,845,1539,object,2448,3264
test_0.jpg,2187,2536,2276,2764,object,2448,3264
test_0.jpg,232,519,361,774,object,2448,3264
test_0.jpg,5,521,225,774,object,2448,3264
test_1.jpg,457,436,574,526,object,1920,2560
test_1.jpg,537,1949,612,2093,object,1920,2560
test_1.jpg,436,2020,534,2095,object,1920,2560
test_1.jpg,1774,1751,1870,1854,object,1920,2560
test_1.jpg,1679,1759,1769,1852,object,1920,2560
test_1.jpg,1578,1762,1674,1852,object,1920,2560
test_1.jpg,1470,1752,1566,1863,object,1920,2560
test_1.jpg,1403,1747,1465,1872,object,1920,2560
test_1.jpg,1231,1764,1330,1869,object,1920,2560
test_1.jpg,1130,1771,1224,1874,object,1920,2560
test_1.jpg,1028,1771,1121,1879,object,1920,2560
test_1.jpg,924,1774,1017,1872,object,1920,2560
test_1.jpg,847,1839,921,1881,object,1920,2560
test_1.jpg,844,1740,919,1833,object,1920,2560
test_1.jpg,759,1839,843,1890,object,1920,2560
test_1.jpg,764,1771,834,1838,object,1920,2560

想要转换的xml格式：

<annotation>
<filename>8a0004.jpg</filename>
<size>
<width>960</width>
<height>1280</height>
<depth>3</depth>
</size>
<object>
<name>object</name>
<bndbox>
<xmin>314</xmin>
<ymin>366</ymin>
<xmax>427</xmax>
<ymax>430</ymax>
</bndbox>
</object>
<object>
<name>object</name>
<bndbox>
<xmin>805</xmin>
<ymin>9</ymin>
<xmax>960</xmax>
<ymax>277</ymax>
</bndbox>
</object>
</annotation>

转换代码：

import csv
import os
from xml.dom import minidom

# 逐行读取csv文件
def create_xml(filename, bboxs,row):

    width = row[6]
    height = row[7]
    depth = 3
    # 1.创建DOM树对象
    dom = minidom.Document()
    # 2.创建根节点。每次都要用DOM对象来创建任何节点。
    root_node = dom.createElement('annotation')
    # 3.用DOM对象添加根节点
    dom.appendChild(root_node)

    filename_node = dom.createElement('filename')
    root_node.appendChild(filename_node)
    # 也用DOM创建文本节点，把文本节点（文字内容）看成子节点
    name_text = dom.createTextNode(filename)
    # 用添加了文本的节点对象（看成文本节点的父节点）添加文本节点
    filename_node.appendChild(name_text)

    # size
    size_node = dom.createElement('size')
    root_node.appendChild(size_node)
    width_node = dom.createElement('width')
    height_node = dom.createElement('height')
    depth_node = dom.createElement('depth')
    # width
    size_node.appendChild(width_node)
    width_text = dom.createTextNode(str(width))
    width_node.appendChild(width_text)
    # height
    size_node.appendChild(height_node)
    height_text = dom.createTextNode(str(height))
    height_node.appendChild(height_text)
    # depth
    size_node.appendChild(depth_node)
    depth_text = dom.createTextNode(str(depth))
    depth_node.appendChild(depth_text)

    for bbox in bboxs:
        # 创建obejct
        object_node = dom.createElement('object')
        root_node.appendChild(object_node)
        # 创建类别name
        name_node = dom.createElement('name')
        name_text = dom.createTextNode('object')
        name_node.appendChild(name_text)
        object_node.appendChild(name_node)
        # 创建bndbox
        # bbox [xmin, ymin, width, height]
        # bbox = ast.literal_eval(bbox)
        xmin, ymin = bbox[0], bbox[1]
        xmax, ymax = bbox[2], bbox[3]

        bndbox = dom.createElement('bndbox')
        object_node.appendChild(bndbox)
        # xmin
        xmin_node = dom.createElement('xmin')
        xmin_text = dom.createTextNode(str(xmin))
        xmin_node.appendChild(xmin_text)
        bndbox.appendChild(xmin_node)
        # ymin
        ymin_node = dom.createElement('ymin')
        ymin_text = dom.createTextNode(str(ymin))
        ymin_node.appendChild(ymin_text)
        bndbox.appendChild(ymin_node)
        # xmax
        xmax_node = dom.createElement('xmax')
        xmax_text = dom.createTextNode(str(xmax))
        xmax_node.appendChild(xmax_text)
        bndbox.appendChild(xmax_node)
        # ymax
        ymax_node = dom.createElement('ymax')
        ymax_text = dom.createTextNode(str(ymax))
        ymax_node.appendChild(ymax_text)
        bndbox.appendChild(ymax_node)

    # 每一个结点对象（包括dom对象本身）都有输出XML内容的方法，如：toxml()--字符串, toprettyxml()--美化树形格式。
    try:
        with open(os.path.join(xml_dir, filename) + '.xml', 'w', encoding='UTF-8') as fh:
            # 4.writexml()第一个参数是目标文件对象，第二个参数是根节点的缩进格式，第三个参数是其他子节点的缩进格式，
            # 第四个参数制定了换行格式，第五个参数制定了xml内容的编码。
            dom.writexml(fh, indent='', addindent='\t', newl='\n', encoding='UTF-8')
            # print('写入xml OK!')
    except Exception as err:
        print('错误信息：{0}'.format(err))


def main():
    with open(csv_filename, 'r', encoding="utf-8") as csvfile:
        reader = csv.reader(csvfile)
        # reader = csv.DictReader(csvfile)
        # 自动获取第一张照片的文件名，并设置为last_image
        last_image = 'test_0.jpg'
        img_num = 1
        bboxs = []
        for row in reader:
            # print(row)
            # print(row[0])
            if row[0] == last_image:
                # 叠加bbox [xmin, ymin, width, height]
                # bboxs.append(row['bbox'])
                box = [row[1], row[2], row[3], row[4]]
                # print(box)
                bboxs.append(box)
            elif row[0] != last_image:
                # 创建xml文件
                create_xml(last_image, bboxs,row)
                last_image = row[0]
                img_num += 1
                # 重置bbox
                bboxs.clear()
                box = [row[1], row[2], row[3], row[4]]
                bboxs.append(box)
        print(img_num)
        print('写入xml OK!')


if __name__ == '__main__':
    # 文件路径
    xml_dir = '/home/hub/wsy/SKU110K_CVPR19/SKU110K/annotations/val_xml'#想要写入的xml文件夹
    csv_filename = os.path.join('/home/hub/wsy/SKU110K_CVPR19/SKU110K/annotations', 'annotations_val.csv')#已经存在的csv路径
    main()

xml转csv的博客：https://blog.csdn.net/poppyty/article/details/115342231