使用selcetive search方法生成伪标签，并写入xml文件

使用selcetive search的方法生成伪标签，写入xml文件

# coding:utf-8
# selective search生成伪标签+修改xml文件
import os
import os.path
import xml.dom.minidom
import xml.etree.ElementTree as et
from PIL import Image
import skimage.data
import matplotlib.pyplot as plt
import matplotlib.patches as mpatches
import selectivesearch
import numpy as np
import cv2
path = 'E:\\Datasets\OWOD\OWOD\VOC2007\Annotations/'
files = os.listdir(path)  # 得到文件夹下所有文件名称
s = []

for xmlFile in files:  # 遍历文件夹
    if not os.path.isdir(xmlFile):  # 判断是否是文件夹，不是文件夹才打开

        label = xmlFile.strip('.xml')
        image_path = 'E:\\Dataset\\VOC2007\\JPEGImages' + '/' + str(label) + '.jpg'
        # image = cv2.imread(image_path)
        img = Image.open(image_path)
        original_image = np.asarray(img)
        H, W = original_image.shape[:2]

        doc = et.parse(os.path.join(path, xmlFile))
        # root is <configuration>
        root_doc = doc.getroot()
        if original_image.shape[-1] != 3:
            doc.write('Annotations2' + '/' + xmlFile)
            print(xmlFile, 'error image')
            continue
        # 生成随机框操作
        img_lbl, regions = selectivesearch.selective_search(original_image, scale=500, sigma=0.9, min_size=200)
        candidates = set()
        for r in regions:
            if r['rect'] in candidates:  # 排除重复的候选区
                continue
            if r['size'] < 2000:  # 排除小于 2000 pixels的候选区域(并不是bounding box中的区域大小)
                continue
            x, y, w, h = r['rect']
            if w / h > 2 or h / w > 2:  # 排除扭曲的候选区域边框  即只保留近似正方形的
                continue
            candidates.add(r['rect'])
        search_region = {}
        i = 0
        for x, y, w, h in candidates:
            norm_x, norm_y, norm_w, norm_h = x / W, y / H, w / W, h / H
            search_region['selective_region_' + str(i)] = (x, y, x + w, y + h)
            i = i + 1
        # new '<selective_region>'
        ns = et.SubElement(root_doc, 'selective_region', attrib={})
        for j in range(len(search_region)):

            # add new '<bndbox>' in <object>
            nb = et.SubElement(ns, 'bndbox', attrib={})
            nxmin = et.SubElement(nb, 'xmin', attrib={})
            nxmin.text = str(search_region['selective_region_' + str(j)][0])

            nymin = et.SubElement(nb, 'ymin', attrib={})
            nymin.text = str(search_region['selective_region_' + str(j)][1])

            nxmax = et.SubElement(nb, 'xmax', attrib={})
            nxmax.text = str(search_region['selective_region_' + str(j)][2])

            nymax = et.SubElement(nb, 'ymax', attrib={})
            nymax.text = str(search_region['selective_region_' + str(j)][3])
            
            et.dump(ns)


        doc.write('Annotations2' + '/' + xmlFile)
        print(xmlFile,'OK')