voc_to_tfrecord

tf_convert_data_main

import tensorflow as tf
import pascalvoc_to_tfrecords

FLAGS = tf.app.flags.FLAGS

tf.app.flags.DEFINE_string(
    'dataset_name', 'pascalvoc',
    'The name of the dataset to convert.')
tf.app.flags.DEFINE_string(
    'dataset_dir', r"C:\Users\ADMIN\Desktop\mty\datasets", #载入数据的地址(主文件夹)
    'Directory where the original dataset is stored.')
tf.app.flags.DEFINE_string(
    'output_name', 'pascalvoc',   #输出文件的名字
    'Basename used for TFRecords output files.')
tf.app.flags.DEFINE_string(
    'output_dir', r"C:\Users\ADMIN\Desktop\mty\datasets\voc",#保存数据的地址
    'Output directory where to store TFRecords files.')


def main(_):
    if not FLAGS.dataset_dir:
        raise ValueError('You must supply the dataset directory with --dataset_dir')
    print('Dataset directory:', FLAGS.dataset_dir)
    print('Output directory:', FLAGS.output_dir)

    if FLAGS.dataset_name == 'pascalvoc':
        pascalvoc_to_tfrecords.run(FLAGS.dataset_dir, FLAGS.output_dir, FLAGS.output_name)
    else:
        raise ValueError('Dataset [%s] was not recognized.' % FLAGS.dataset_name)

if __name__ == '__main__':
    tf.app.run()

pascalvoc_to_tfrecords

import os
import sys
import random

import numpy as np
import tensorflow as tf

import xml.etree.ElementTree as ET

from dataset_utils import int64_feature, float_feature, bytes_feature
from pascalvoc_common import VOC_LABELS

# Original dataset organisation.
# 给定xml标注数据所在的文件夹名称(文件夹中必须包含Annotations，JPEGImages这两个子文件夹)
DIRECTORY_ANNOTATIONS = 'Annotations/'
# 给定图像数据所在的文件夹名称
DIRECTORY_IMAGES = 'JPEGImages/'

# TFRecords convertion parameters.
RANDOM_SEED = 4242
# 给定每个tfrecord文件存储多少原始图像的数据
SAMPLES_PER_FILES = 2


def _process_image(directory, name):
    """Process a image and annotation file.

    Args:
      filename: string, path to an image file e.g., '/path/to/example.JPG'.
      coder: instance of ImageCoder to provide TensorFlow image coding utils.
    Returns:
      image_buffer: string, JPEG encoding of RGB image.
      height: integer, image height in pixels.
      width: integer, image width in pixels.
    """
    # Read the image file.
    # 1. 构造图像路径
    filename = os.path.join(directory, DIRECTORY_IMAGES, name + '.jpg')


    # 2. 加载数据
    image_data = tf.gfile.FastGFile(filename, 'rb').read()

    # Read the XML annotation file.
    # 3. 构造xml文件路径
    filename = os.path.join(directory, DIRECTORY_ANNOTATIONS, name + '.xml')

    # 4. xml文件解析
    tree = ET.parse(filename)
    # a. 得到xml文件对应的根节点
    root = tree.getroot()

    # Image shape.
    # b. 获取图像大小信息
    # 从根节点中获取第一个size子标签
    size = root.find('size')
    shape = [int(size.find('height').text),
             int(size.find('width').text),
             int(size.find('depth').text)]
    # Find annotations.
    # c. 获取边框信息
    bboxes = []
    labels = []
    labels_text = []
    difficult = []
    truncated = []
    # 从根节点中获取所有object标签的值(返回的是一个列表)
    for obj in root.findall('object'):
        # a. 获取当前物体的所属类别名称
        label = obj.find('name').text
        # b. 根据标签，将标签转换为索引id和byte数组，并添加集合中
        labels.append(int(VOC_LABELS[label][0])) #注意这边VOC_LABELS必须和用标注工具标注时候的label相同
        labels_text.append(label.encode('ascii'))

        # c. 获取difficult标记信息，用于判断当前这个物体是否属于比较难预测的物体
        if obj.find('difficult'):
            difficult.append(int(obj.find('difficult').text))
        else:
            # 0表示预测不难
            difficult.append(0)