PASCAL VOC data set is converted to yolo data set format

1. PASCAL VOC data format

<?xml version='1.0' encoding='utf-8'?>
<annotation verified="no">
  <folder>JPEGImages</folder>
  <filename>2018_06_05_09_06_55_065</filename>
  <path>F:\receive\VOC2007\JPEGImages\2018_06_05_09_06_55_065.jpg</path>
  <source>
    <database>Unknown</database>
  </source>
  <size>
    <width>2048</width>
    <height>1536</height>
    <depth>3</depth>
  </size>
  <segmented>0</segmented>
  <object>
    <name>1</name>
    <pose>Unspecified</pose>
    <truncated>0</truncated>
    <difficult>0</difficult>
    <bndbox>
      <xmin>530</xmin>
      <ymin>752</ymin>
      <xmax>1498</xmax>
      <ymax>1326</ymax>
    </bndbox>
  </object>
</annotation>

2. yolo data format

    class x_center y_center w h The coordinate position and the width and height of the box are divided by the width and height of the picture

0 0.947510 0.546224 0.049316 0.050781
0 0.434326 0.586263 0.217285 0.194661

3. The file layout is shown in the figure

      Images are stored in images, labels_voc are stored in *.xml (labels), and labels are stored in *.txt (converted labels)

       Classes.names stores the name of each category, one category name per line

       The contents of pascalVOC2yolov3.py are as follows:

#coding:utf-8
from __future__ import print_function

import os
import random
import glob
import xml.etree.ElementTree as ET

def xml_reader(filename):
    """ Parse a PASCAL VOC xml file """
    tree = ET.parse(filename)
    size = tree.find('size')
    width = int(size.find('width').text)
    height = int(size.find('height').text)
    objects = []
    for obj in tree.findall('object'):
        obj_struct = {}
        obj_struct['name'] = obj.find('name').text
        bbox = obj.find('bndbox')
        obj_struct['bbox'] = [int(bbox.find('xmin').text),
                              int(bbox.find('ymin').text),
                              int(bbox.find('xmax').text),
                              int(bbox.find('ymax').text)]
        objects.append(obj_struct)
    return width, height, objects


def voc2yolo(filename):
    classes_dict = {}
    with open("classes.names") as f:
        for idx, line in enumerate(f.readlines()):
            class_name = line.strip()
            classes_dict[class_name] = idx
    
    width, height, objects = xml_reader(filename)

    lines = []
    for obj in objects:
        x, y, x2, y2 = obj['bbox']
        class_name = obj['name']
        label = classes_dict[class_name]
        cx = (x2+x)*0.5 / width
        cy = (y2+y)*0.5 / height
        w = (x2-x)*1. / width
        h = (y2-y)*1. / height
        line = "%s %.6f %.6f %.6f %.6f\n" % (label, cx, cy, w, h)
        lines.append(line)

    txt_name = filename.replace(".xml", ".txt").replace("labels_voc", "labels")
    with open(txt_name, "w") as f:
        f.writelines(lines)


def get_image_list(image_dir, suffix=['jpg', 'jpeg', 'JPG', 'JPEG','png']):
    '''get all image path ends with suffix'''
    if not os.path.exists(image_dir):
        print("PATH:%s not exists" % image_dir)
        return []
    imglist = []
    for root, sdirs, files in os.walk(image_dir):
        if not files:
            continue
        for filename in files:
            filepath = "data/custom/" + os.path.join(root, filename) + "\n"
            if filename.split('.')[-1] in suffix:
                imglist.append(filepath)
    return imglist


def imglist2file(imglist):
    random.shuffle(imglist)
    train_list = imglist[:-100]
    valid_list = imglist[-100:]
    with open("train.txt", "w") as f:
        f.writelines(train_list)
    with open("valid.txt", "w") as f:
        f.writelines(valid_list)


if __name__ == "__main__":
    xml_path_list = glob.glob("labels_voc/*.xml")
    for xml_path in xml_path_list:
        voc2yolo(xml_path)


    imglist = get_image_list("images")
    imglist2file(imglist)

        

After the data conversion is completed, the detection model can be trained using the yolo series of algorithms.