【tensorflow 深度学习】8.训练图片分类模型

1.训练图片分类模型的三种方法

    （1）.从无到有，先确定好算法框架，准备好需要训练的数据集，从头开始训练，参数一开始也是初始化的随机值，一个批次一个批次地进行训练。

    （2）.准备好已经训练好的模型，权值参数也都已经确定，只训练最后一层，因为前面的参数都是经过大量图片的训练来的，所以参数都比较好，比如卷积层主要的作用的对图像特征的提取，我们要做自己的分类模型的话也得对图像进行特征提取，做特征提取的话直接使用训练好的权值也行。

    （3）.跟2差不多，不同的是都之前的参数也做微调。

2.retrain图片分类模型

    (1). https://github.com/tensorflow/tensorflow 下载官方包 有一些官方提供的案例，里面有后面要用到的retrain.py文件。

    (2).下载图片集

    网址：http://www.robots.ox.ac.uk/~vgg/data/

    （3）然后写批处理文件

    ：

activate py3 ^
python E:/graduate_student/deep_learning/tensorflow-master/tensorflow-master/tensorflow/examples/image_retraining/retrain.py ^
--bottleneck_dir bottleneck ^
--how_many_training_steps 200 ^
--model_dir D:/software/mycodes/python35/py3/inception_model/ ^
--output_graph output_graph.pb ^
--output_labels output_labels.txt ^
--image_dir imagedata/
pause

    bottleneck 瓶颈，图片预处理的时候把这个值算出来

    output_graph 输出训练好的模型到当前文件夹

    output_labels 输出训练好的标签到当前文件夹 

    image的格式需要里面文件夹的名字代表分类类型。（里面图片不能由大写字母也不能由中文！！）

    运行批处理文件，得到：

    （4）测试训练好的模型：

    代码：

# coding: utf-8

# In[1]:

import tensorflow as tf
import os
import numpy as np
import re
from PIL import Image
import matplotlib.pyplot as plt


# In[2]:

lines = tf.gfile.GFile('E:/graduate_student/deep_learning/a-tensorflow/9/retain/output_labels.txt').readlines()
uid_to_human = {}
#一行一行读取数据
for uid,line in enumerate(lines) :
    #去掉换行符
    line=line.strip('\n')
    uid_to_human[uid] = line

def id_to_string(node_id):
    if node_id not in uid_to_human:
        return ''
    return uid_to_human[node_id]


#创建一个图来存放google训练好的模型
with tf.gfile.FastGFile('E:/graduate_student/deep_learning/a-tensorflow/9/retain/output_graph.pb', 'rb') as f:
    graph_def = tf.GraphDef()
    graph_def.ParseFromString(f.read())
    tf.import_graph_def(graph_def, name='')


with tf.Session() as sess:
    softmax_tensor = sess.graph.get_tensor_by_name('final_result:0')
    #遍历目录
    for root,dirs,files in os.walk('E:/graduate_student/deep_learning/a-tensorflow/9/retain/images/'):
        for file in files:
            #载入图片
            image_data = tf.gfile.FastGFile(os.path.join(root,file), 'rb').read()
            predictions = sess.run(softmax_tensor,{'DecodeJpeg/contents:0': image_data})#图片格式是jpg格式
            predictions = np.squeeze(predictions)#把结果转为1维数据

            #打印图片路径及名称
            image_path = os.path.join(root,file)
            print(image_path)
            #显示图片
            img=Image.open(image_path)
            plt.imshow(img)
            plt.axis('off')
            plt.show()

            #排序
            top_k = predictions.argsort()[::-1]
            print(top_k)
            for node_id in top_k:     
                #获取分类名称
                human_string = id_to_string(node_id)
                #获取该分类的置信度
                score = predictions[node_id]
                print('%s (score = %.5f)' % (human_string, score))
            print()

结果：

    使用finetune的优点：1.训练速度快，计算量少，只计算最后一层。2.迭代周期少，因为训练的权值少。3.需要使用到图片的数据量比较少。

3.重头开始训练图片识别模型

    （1）tensorflow官方包里下载：https://github.com/tensorflow/models

    （2）准备好分类图片

    （3）图像预处理，生成tfrecord文件。

    程序：

# coding: utf-8

# In[2]:

import tensorflow as tf
import os
import random
import math
import sys


# In[3]:

#验证集数量
_NUM_TEST = 500
#随机种子
_RANDOM_SEED = 0
#数据块
_NUM_SHARDS = 5
#数据集路径
DATASET_DIR = "E:/graduate_student/deep_learning/a-tensorflow\9/retain/imagedata/"
#标签文件名字
LABELS_FILENAME = "E:/graduate_student/deep_learning/a-tensorflow/9/retain/labels.txt"

#定义tfrecord文件的路径+名字
def _get_dataset_filename(dataset_dir, split_name, shard_id):
    output_filename = 'image_%s_%05d-of-%05d.tfrecord' % (split_name, shard_id, _NUM_SHARDS)
    return os.path.join(dataset_dir, output_filename)

#判断tfrecord文件是否存在
def _dataset_exists(dataset_dir):
    for split_name in ['train', 'test']:
        for shard_id in range(_NUM_SHARDS):
            #定义tfrecord文件的路径+名字
            output_filename = _get_dataset_filename(dataset_dir, split_name, shard_id)
        if not tf.gfile.Exists(output_filename):
            return False
    return True

#获取所有文件以及分类
def _get_filenames_and_classes(dataset_dir):
    #数据目录
    directories = []
    #分类名称
    class_names = []
    for filename in os.listdir(dataset_dir):
        #合并文件路径
        path = os.path.join(dataset_dir, filename)
        #判断该路径是否为目录
        if os.path.isdir(path):
            #加入数据目录
            directories.append(path)
            #加入类别名称
            class_names.append(filename)

    photo_filenames = []
    #循环每个分类的文件夹
    for directory in directories:
        for filename in os.listdir(directory):
            path = os.path.join(directory, filename)
            #把图片加入图片列表
            photo_filenames.append(path)

    return photo_filenames, class_names

def int64_feature(values):
    if not isinstance(values, (tuple, list)):
        values = [values]
    return tf.train.Feature(int64_list=tf.train.Int64List(value=values))

def bytes_feature(values):
    return tf.train.Feature(bytes_list=tf.train.BytesList(value=[values]))

def image_to_tfexample(image_data, image_format, class_id):
    #Abstract base class for protocol messages.
    return tf.train.Example(features=tf.train.Features(feature={
      'image/encoded': bytes_feature(image_data),
      'image/format': bytes_feature(image_format),
      'image/class/label': int64_feature(class_id),
    }))

def write_label_file(labels_to_class_names, dataset_dir,filename=LABELS_FILENAME):
    labels_filename = os.path.join(dataset_dir, filename)
    with tf.gfile.Open(labels_filename, 'w') as f:
        for label in labels_to_class_names:
            class_name = labels_to_class_names[label]
            f.write('%d:%s\n' % (label, class_name))

#把数据转为TFRecord格式
def _convert_dataset(split_name, filenames, class_names_to_ids, dataset_dir):
    assert split_name in ['train', 'test']
    #计算每个数据块有多少数据
    num_per_shard = int(len(filenames) / _NUM_SHARDS)
    with tf.Graph().as_default():
        with tf.Session() as sess:
            for shard_id in range(_NUM_SHARDS):
                #定义tfrecord文件的路径+名字
                output_filename = _get_dataset_filename(dataset_dir, split_name, shard_id)
                with tf.python_io.TFRecordWriter(output_filename) as tfrecord_writer:
                    #每一个数据块开始的位置
                    start_ndx = shard_id * num_per_shard
                    #每一个数据块最后的位置
                    end_ndx = min((shard_id+1) * num_per_shard, len(filenames))
                    for i in range(start_ndx, end_ndx):
                        try:
                            sys.stdout.write('\r>> Converting image %d/%d shard %d' % (i+1, len(filenames), shard_id))
                            sys.stdout.flush()
                            #读取图片
                            image_data = tf.gfile.FastGFile(filenames[i], 'r').read()
                            #获得图片的类别名称
                            class_name = os.path.basename(os.path.dirname(filenames[i]))
                            #找到类别名称对应的id
                            class_id = class_names_to_ids[class_name]
                            #生成tfrecord文件
                            example = image_to_tfexample(image_data, b'jpg', class_id)
                            tfrecord_writer.write(example.SerializeToString())
                        except IOError as e:
                            print("Could not read:",filenames[i])
                            print("Error:",e)
                            print("Skip it\n")
                            
    sys.stdout.write('\n')
    sys.stdout.flush()


if __name__ == '__main__':
    #判断tfrecord文件是否存在
    if _dataset_exists(DATASET_DIR):
        print('tfcecord文件已存在')
    else:
        #获得所有图片以及分类
        photo_filenames, class_names = _get_filenames_and_classes(DATASET_DIR)
        #把分类转为字典格式，类似于{'house': 3, 'flower': 1, 'plane': 4, 'guitar': 2, 'animal': 0}
        class_names_to_ids = dict(zip(class_names, range(len(class_names))))

        #把数据切分为训练集和测试集
        random.seed(_RANDOM_SEED)
        random.shuffle(photo_filenames)
        training_filenames = photo_filenames[_NUM_TEST:]
        testing_filenames = photo_filenames[:_NUM_TEST]

        #数据转换
        _convert_dataset('train', training_filenames, class_names_to_ids, DATASET_DIR)
        _convert_dataset('test', testing_filenames, class_names_to_ids, DATASET_DIR)

        #输出labels文件
        labels_to_class_names = dict(zip(range(len(class_names)), class_names))
        write_label_file(labels_to_class_names, DATASET_DIR)

得到:

和

    （4）批处理文件：

        1.在slim里面加入image文件夹，里面放入图片tfrecord文件

        2.在slim文件夹里的datasets文件夹里新建

        程序如下：

"""Provides data for the flowers dataset.

The dataset scripts used to create the dataset can be found at:
tensorflow/models/slim/datasets/download_and_convert_flowers.py
"""

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import os
import tensorflow as tf

from datasets import dataset_utils

slim = tf.contrib.slim

_FILE_PATTERN = 'image_%s_*.tfrecord'

SPLITS_TO_SIZES = {'train': 1000, 'test': 500}

_NUM_CLASSES = 5

_ITEMS_TO_DESCRIPTIONS = {
    'image': 'A color image of varying size.',
    'label': 'A single integer between 0 and 4',
}


def get_split(split_name, dataset_dir, file_pattern=None, reader=None):
  """Gets a dataset tuple with instructions for reading flowers.

  Args:
    split_name: A train/validation split name.
    dataset_dir: The base directory of the dataset sources.
    file_pattern: The file pattern to use when matching the dataset sources.
      It is assumed that the pattern contains a '%s' string so that the split
      name can be inserted.
    reader: The TensorFlow reader type.

  Returns:
    A `Dataset` namedtuple.

  Raises:
    ValueError: if `split_name` is not a valid train/validation split.
  """
  if split_name not in SPLITS_TO_SIZES:
    raise ValueError('split name %s was not recognized.' % split_name)

  if not file_pattern:
    file_pattern = _FILE_PATTERN
  file_pattern = os.path.join(dataset_dir, file_pattern % split_name)

  # Allowing None in the signature so that dataset_factory can use the default.
  if reader is None:
    reader = tf.TFRecordReader

  keys_to_features = {
      'image/encoded': tf.FixedLenFeature((), tf.string, default_value=''),
      'image/format': tf.FixedLenFeature((), tf.string, default_value='png'),
      'image/class/label': tf.FixedLenFeature(
          [], tf.int64, default_value=tf.zeros([], dtype=tf.int64)),
  }

  items_to_handlers = {
      'image': slim.tfexample_decoder.Image(),
      'label': slim.tfexample_decoder.Tensor('image/class/label'),
  }

  decoder = slim.tfexample_decoder.TFExampleDecoder(
      keys_to_features, items_to_handlers)

  labels_to_names = None
  if dataset_utils.has_labels(dataset_dir):
    labels_to_names = dataset_utils.read_label_file(dataset_dir)

  return slim.dataset.Dataset(
      data_sources=file_pattern,
      reader=reader,
      decoder=decoder,
      num_samples=SPLITS_TO_SIZES[split_name],
      items_to_descriptions=_ITEMS_TO_DESCRIPTIONS,
      num_classes=_NUM_CLASSES,
      labels_to_names=labels_to_names)

        然后在datasets文件夹里的里修改：

        添加myimages。

           3.在slim目录下新建批处理文件。 其中，train_image_classifier.py是在（1）里下载的文件里的程序，在slim文件夹里面，slim文件夹需要拷贝到当前目录。

D:\Anaconda2\envs\PY3\python E:/graduate_student/deep_learning/models-master/models-master/research/slim/train_image_classifier.py ^
--train_dir=D:/software/mycodes/python3/python35/captcha/model/ ^
--dataset_name=myimages ^
--dataset_split_name=train ^
--dataset_dir=D:/software/mycodes/python3/python35/captcha/image ^
--batch_size=10 ^
--max_number_of_steps=10000 ^
--model_name=inception_v3 ^
pause

    然后运行批处理文件。缓慢运行。