【目标检测】图像数据增强——翻转及修改对应的xml标注

Syntax

cv2.flip(src, flipCode[, dst])

src: 原始图像矩阵，dst: 变换后的矩阵，flipMode: 翻转模式，有三种模式
args

flipCode	annotation
1	水平翻转
0	垂直翻转
-1	水平垂直翻转

代码
三种翻转方式可以一起进行，也可以按照需要注释代码选择一种翻转方式。
该代码可以同时修改xml中bndbox和filename以及自身的名字。

import cv2
import os
import xml.etree.ElementTree as ET

# 水平镜像翻转
def h_MirrorImg(img_path,img_write_path):
    img = cv2.imread(img_path)
    mirror_img = cv2.flip(img, 1)  
    cv2.imwrite(img_write_path,mirror_img)

# 垂直翻转
def v_MirrorImg(img_path,img_write_path):
    img = cv2.imread(img_path)
    mirror_img = cv2.flip(img, 0)  
    cv2.imwrite(img_write_path,mirror_img)

# 水平垂直翻转
def a_MirrorImg(img_path,img_write_path):
    img = cv2.imread(img_path)
    mirror_img = cv2.flip(img, -1)  
    cv2.imwrite(img_write_path,mirror_img)


def h_MirrorAnno(anno_path,anno_write_path,name):
    tree = ET.parse(anno_path)
    root = tree.getroot()
    size=root.find('size')
    w=int(size.find('width').text)
    objects = root.findall("object")
    for obj in objects:
        bbox = obj.find('bndbox')
        x1 = float(bbox.find('xmin').text)
        x2 = float(bbox.find('xmax').text)
        x1=w-x1+1
        x2=w-x2+1

        assert x1>0
        assert x2>0

        bbox.find('xmin').text=str(int(x2))
        bbox.find('xmax').text=str(int(x1))
        
    for node in root.findall('filename'):
        if node != name:
            node.text = name

    tree.write(anno_write_path)  # 保存修改后的XML文件


def v_MirrorAnno(anno_path,anno_write_path,name):
    tree = ET.parse(anno_path)
    root = tree.getroot()
    size = root.find('size')
    h=int(size.find('height').text)
    objects = root.findall("object")
    for obj in objects:
        bbox = obj.find('bndbox')
        y1 = float(bbox.find('ymin').text)
        y2 = float(bbox.find('ymax').text)

        y1=h-y1+1
        y2=h-y2+1

        assert y1>0
        assert y2>0

        bbox.find('ymin').text=str(int(y2))
        bbox.find('ymax').text=str(int(y1))
        
    for node in root.findall('filename'):
        if node != name:
            node.text = name

    tree.write(anno_write_path)  # 保存修改后的XML文件


def a_MirrorAnno(anno_path,anno_write_path,name):
    tree = ET.parse(anno_path)
    root = tree.getroot()
    size = root.find('size')
    w=int(size.find('width').text)
    h = int(size.find('height').text)
    objects = root.findall("object")
    for obj in objects:
        bbox = obj.find('bndbox')
        x1 = float(bbox.find('xmin').text)
        y1 = float(bbox.find('ymin').text)
        x2 = float(bbox.find('xmax').text)
        y2 = float(bbox.find('ymax').text)

        x1=w-x1+1
        x2=w-x2+1

        y1 = h - y1+1
        y2 = h - y2+1

        assert x1 > 0
        assert x2 > 0
        assert y1 > 0
        assert y2 > 0

        bbox.find('xmin').text=str(int(x2))
        bbox.find('xmax').text=str(int(x1))
        bbox.find('ymin').text=str(int(y2))
        bbox.find('ymax').text=str(int(y1))
        
    for node in root.findall('filename'):
        if node != name:
            node.text = name

    tree.write(anno_write_path)  # 保存修改后的XML文件


def mirror(img_dir, xml_dir, img_write_dir, xml_write_dir):
    if not os.path.exists(img_write_dir):
        os.makedirs(img_write_dir)
    if not os.path.exists(xml_write_dir):
        os.makedirs(xml_write_dir)

    img_name = os.listdir(img_dir)
    for img in img_name:
        img_path = os.path.join(img_dir,img)
        # 注意img[:-4]，如果后缀是jpeg则改成img[:-5]
        h_img_write_path=os.path.join(img_write_dir,img[:-4]+'_hflip'+'.jpg')
        anno_path=os.path.join(xml_dir,img[:-4]+'.xml')
        h_anno_write_path = os.path.join(xml_write_dir, img[:-4]+'_hflip'+'.xml')
        #
        v_img_write_path = os.path.join(img_write_dir, img[:-4] + '_vflip' + '.jpg')
        v_anno_write_path = os.path.join(xml_write_dir, img[:-4] + '_vflip' + '.xml')
        #
        a_img_write_path = os.path.join(img_write_dir, img[:-4] + '_aflip' + '.jpg')
        a_anno_write_path = os.path.join(xml_write_dir, img[:-4] + '_aflip' + '.xml')
        #
        h_MirrorImg(img_path,h_img_write_path)
        v_MirrorImg(img_path,v_img_write_path)
        a_MirrorImg(img_path, a_img_write_path)
        h_MirrorAnno(anno_path, h_anno_write_path, img[:-4]+'_hflip'+'.jpg')
        v_MirrorAnno(anno_path, v_anno_write_path, img[:-4]+'_vflip'+'.jpg')
        a_MirrorAnno(anno_path, a_anno_write_path, img[:-4]+'_aflip'+'.jpg')


if __name__ == '__main__':
    img_path = './dog'  # 图片文件夹路径
    xml_path = './dog_xml'  # xml标注文件夹路径
    img_write_path = './dog_rotatedimg'  # 翻转后的图片保存路径
    xml_write_path = './dog_rotatedxml'  # 修改后的xml标注保存路径
    mirror(img_path, xml_path, img_write_path, xml_write_path)

结果示例
原图：

Flipped Horizontally 水平翻转：
Flipped Vertically 垂直翻转：
Flipped Horizontally & Vertically 水平垂直翻转：

参考：
目标检测数据集的增强（旋转，镜像，亮度等）含针对VOC标注格式数据的源码