TensorFlow使用inception模型进行flower识别训练+修改原始inception实现predict

这是我对原始Inception做了修改之后传到github上的github链接

主要的修改是实现了输出filename和对应的label 并提供了运行相关py文件的运行命令

有需要的可以去下载

自己尝试运行inception 中的flowers_train.py 在此过程中 遇到了一些坑 在网上也没有找到详细讲述如何运行的博客

在这里写下来与大家分享我作为一个小白遇到的问题和解决的办法

TensorFlow提供了很多模型代码models 下载到本地解压  只使用其中的inception

用PyCharm打开 运行flowers_train.py（可能需要修改编译环境 选择tensorflow）

此时报错 大概就是提示inception.XXX 用不了

此时需要在第二层inception文件夹中新建空的py文件 “__init__.py”

同样的还会提示slim.XXX 用不了 在slim文件夹中新建空的py文件 “__init__.py”

再运行flowers_train.py 还是无法运行  提示找不到数据

因为我用的mac 于是就选择使用inception/inception/data/download_and_preprocess_flowers_mac.sh

将这个.sh文件拖到命令行窗口中 后面需要加上保存文件的路径 然后执行

执行过程中 一开始是没问题的 下载flower数据（此过程中最好打开 vpn）

但是下载完毕之后呢 根据.sh命令 需要将数据分为 train和validation 

然后我这就报错了 大概的意思就是gshuf commound无法执行

这个解决方案百度吧 就是安装一个东西就可以了。。。

同时呢 还有个报错 就是找不到build_image_data

这个呢 我一开始以为是需要添加一个build_image_data文件夹或者文件 发现不对

后来发现有一个build_image_data.py的文件 就按照路径要求 把这个文件复制过去了 发现还是有点问题

于是呢 看到了这个解决方法 于是就把里面的shell文件内容 复制过来 执行 得到一个输出

输出的内容有关于build_image_data.py的执行命令 于是就在terminal中 进入到build_image_data.py

所在的路径 运行指令 就可以将下载的图片转换为TFRecord 

以下我修改后的shell文件内容

#!/bin/bash
# Copyright 2016 Google Inc. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================

# Script to download and preprocess the flowers data set. This data set
# provides a demonstration for how to perform fine-tuning (i.e. tranfer
# learning) from one model to a new data set.
#
# This script provides a demonstration for how to prepare an arbitrary
# data set for training an Inception v3 model.
#
# We demonstrate this with the flowers data set which consists of images
# of labeled flower images from 5 classes:
#
# daisy, dandelion, roses, sunflowers, tulips
#
# The final output of this script are sharded TFRecord files containing
# serialized Example protocol buffers. See build_image_data.py for
# details of how the Example protocol buffer contains image data.
#
# usage:
#  ./download_and_preprocess_flowers.sh [data-dir]
set -e

if [ -z "$1" ]; then
  echo "Usage: download_and_preprocess_flowers.sh [data dir]"
  exit
fi

# Create the output and temporary directories.
DATA_DIR="${1%/}"
SCRATCH_DIR="${DATA_DIR}/raw-data/"
mkdir -p "${DATA_DIR}"
mkdir -p "${SCRATCH_DIR}"
WORK_DIR="$0.runfiles/inception/inception"

# Download the flowers data.
DATA_URL="http://download.tensorflow.org/example_images/flower_photos.tgz"
CURRENT_DIR=$(pwd)
cd "${DATA_DIR}"
TARBALL="flower_photos.tgz"
if [ ! -f ${TARBALL} ]; then
  echo "Downloading flower data set."
  curl -o ${TARBALL} "${DATA_URL}"
else
  echo "Skipping download of flower data."
fi

#echo ${WORK_DIR}
#/Users/youngkl/Desktop/inception/inception/tmp
#echo ${DATA_DIR}
#/Users/youngkl/Desktop/inception/inception/tmp

# Note the locations of the train and validation data.
TRAIN_DIRECTORY="${SCRATCH_DIR}train/"
VALIDATION_DIRECTORY="${SCRATCH_DIR}validation/"

# Expands the data into the flower_photos/ directory and rename it as the
# train directory.
tar xf flower_photos.tgz
rm -rf "${TRAIN_DIRECTORY}" "${VALIDATION_DIRECTORY}"
mv flower_photos "${TRAIN_DIRECTORY}"

# Generate a list of 5 labels: daisy, dandelion, roses, sunflowers, tulips
LABELS_FILE="${SCRATCH_DIR}/labels.txt"
ls -1 "${TRAIN_DIRECTORY}" | grep -v 'LICENSE' | sed 's/\///' | sort > "${LABELS_FILE}"

# Generate the validation data set.
while read LABEL; do
  VALIDATION_DIR_FOR_LABEL="${VALIDATION_DIRECTORY}${LABEL}"
  TRAIN_DIR_FOR_LABEL="${TRAIN_DIRECTORY}${LABEL}"

  # Move the first randomly selected 100 images to the validation set.
  mkdir -p "${VALIDATION_DIR_FOR_LABEL}"
  VALIDATION_IMAGES=$(ls -1 "${TRAIN_DIR_FOR_LABEL}" | gshuf | head -100)
  for IMAGE in ${VALIDATION_IMAGES}; do
    mv -f "${TRAIN_DIRECTORY}${LABEL}/${IMAGE}" "${VALIDATION_DIR_FOR_LABEL}"
  done
done < "${LABELS_FILE}"

# Build the TFRecords version of the image data.
cd "${CURRENT_DIR}"
BUILD_SCRIPT="${WORK_DIR}/build_image_data"
OUTPUT_DIRECTORY="${DATA_DIR}"
echo "${BUILD_SCRIPT}"
echo "${CURRENT_DIR}"

echo "python build_image_data.py  --train_directory=${TRAIN_DIRECTORY}  --validation_directory=${VALIDATION_DIRECTORY}  --output_directory=${OUTPUT_DIRECTORY} --labels_file=${LABELS_FILE}"

数据有了之后 继续运行flowers_train.py 发现还是有问题  大概还是找不到数据的提示信息

大概想到 这个py文件执行是需要输入参数的 

查看点击打开链接 中间部分   How to Retrain a Trained Model on the Flowers Data

在这之下 发现有与之前出现的.sh文件有类似的部分 想到同样的办法 把参数具体内容输出

于是我 新建了一个flower.sh文件  文件内容如下

cd "${CURRENT_DIR}"

# Directory where the flowers data resides.
FLOWERS_DATA_DIR=/Users/youngkl/Desktop/inception/inception/tmp/raw-data/

# Directory where to save the checkpoint and events files.
TRAIN_DIR=/Users/youngkl/Desktop/inception/inception/tmp/


echo "python flowers_train.py  --train_directory=${TRAIN_DIR}   --data_dir=${FLOWERS_DATA_DIR}   --fine_tune=False   --initial_learning_rate=0.001   --input_queue_memory_factor=1"

#python flowers_train.py  --train_directory=/Users/youngkl/Desktop/inception/inception/tmp/   --data_dir=/Users/youngkl/Desktop/inception/inception/tmp/raw-data/   --fine_tune=False   --initial_learning_rate=0.001   --input_queue_memory_factor=1

注意 需要对其中的DIR值做修改 将这个.sh文件拖到terminal 回车执行 输出了想要的内容

复制 在terminal中 进入到flowers_train.py所在的文件目录 执行就可以了

python flowers_train.py  --train_directory=/Users/youngkl/Desktop/inception/inception/tmp/   --data_dir=/Users/youngkl/Desktop/inception/inception/tmp/raw-data/   --fine_tune=False   --initial_learning_rate=0.001   --input_queue_memory_factor=1

之后发现在这个命令中还可以添加其他的内容 比如gpu数量 最多迭代的次数 还可以设置使用预先训练好的模型进一步调节  

具体的都可以在inception_train.py中进行查看

以上是训练阶段遇到的问题 后来在测试阶段又有些问题了

一开始validation的时候呢 是没有问题的 可以输出top1 precision和top5 recall 问题出现在test阶段

在test阶段 我想输出文件名对应的label

但是原始的inception_eval.py的_eval_once函数里 并没有输出 所以我们需要进行修改

首先呢 原始的evaluate函数里面 得到了labels和logits值 将这个值传到_eval_once进行输出即可

注意 tensorflow中  logits值需要用tf.nn.softmax 才得到网络对每个类别预测的概率  概率最大的id就是预测的类别（数组中id是从0开始算的）

但是预测时输入很多图片 没办法得到输出的类别对应的哪个文件 此时需要输出文件名 但是原始的程序里面是没有的 所以需要做修改

这里给出修改的代码

inception_eval.py

# Copyright 2016 Google Inc. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""A library to evaluate Inception on a single GPU.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

from datetime import datetime
import math
import os.path
import time


import numpy as np
import tensorflow as tf

from inception import image_processing
from inception import inception_model as inception


FLAGS = tf.app.flags.FLAGS

tf.app.flags.DEFINE_string('eval_dir', '/tmp/imagenet_eval',
                           """Directory where to write event logs.""")
tf.app.flags.DEFINE_string('checkpoint_dir', '/home/yangkunlin/home/',
                           """Directory where to read model checkpoints.""")

# Flags governing the frequency of the eval.
tf.app.flags.DEFINE_integer('eval_interval_secs', 60 * 5,
                            """How often to run the eval.""")
tf.app.flags.DEFINE_boolean('run_once', False,
                            """Whether to run eval only once.""")

# Flags governing the data used for the eval.
tf.app.flags.DEFINE_integer('num_examples', 2813,
                            """Number of examples to run. Note that the eval """
                            """ImageNet dataset contains 50000 examples.""")
tf.app.flags.DEFINE_string('subset', 'validation',
                           """Either 'validation' or 'train'.""")


def _eval_once(saver, summary_writer, filenames, logits, labels, top_1_op, top_5_op, summary_op):
  """Runs Eval once.

  Args:
    saver: Saver.
    summary_writer: Summary writer.
    top_1_op: Top 1 op.
    top_5_op: Top 5 op.
    summary_op: Summary op.
  """
  print ("path")
  print (FLAGS.checkpoint_dir)
  with tf.Session() as sess:
    ckpt = tf.train.get_checkpoint_state(FLAGS.checkpoint_dir)
    if ckpt and ckpt.model_checkpoint_path:
      if os.path.isabs(ckpt.model_checkpoint_path):
        # Restores from checkpoint with absolute path.
        saver.restore(sess, ckpt.model_checkpoint_path)
      else:
        # Restores from checkpoint with relative path.
        saver.restore(sess, os.path.join(FLAGS.checkpoint_dir,
                                         ckpt.model_checkpoint_path))

      # Assuming model_checkpoint_path looks something like:
      #   /my-favorite-path/imagenet_train/model.ckpt-0,
      # extract global_step from it.
      global_step = ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1]
      print('Successfully loaded model from %s at step=%s.' %
            (ckpt.model_checkpoint_path, global_step))
    else:
      print('No checkpoint file found')
      return

    # Start the queue runners.
    coord = tf.train.Coordinator()
    try:
      threads = []
      for qr in tf.get_collection(tf.GraphKeys.QUEUE_RUNNERS):
        threads.extend(qr.create_threads(sess, coord=coord, daemon=True,
                                         start=True))

      num_iter = int(math.ceil(FLAGS.num_examples / FLAGS.batch_size))
      # Counts the number of correct predictions.
      count_top_1 = 0.0
      count_top_5 = 0.0
      total_sample_count = num_iter * FLAGS.batch_size
      step = 0

      print('%s: starting evaluation on (%s).' % (datetime.now(), FLAGS.subset))
      start_time = time.time()
      while step < num_iter and not coord.should_stop():
        filenames_,logits_, labels_, top_1, top_5 = sess.run([filenames,logits, labels, top_1_op, top_5_op])
        # print (tf.nn.softmax(logits_).eval())
        print (filenames_)
        logi = tf.nn.softmax(logits_).eval()
        # print (labels_)
        # print (logits_.shape[0])
        # print (logits_.shape[1])
        row = logits_.shape[0]
        col = logits_.shape[1]
        for i in range(row):
          # print (filenames[i].eval())
          x = -1.0
          id = -1
          for j in range(col):
            if logi[i][j] > x:
              x = logi[i][j]
              id = j
          print (id)
        count_top_1 += np.sum(top_1)
        count_top_5 += np.sum(top_5)
        step += 1
        if step % 20 == 0:
          duration = time.time() - start_time
          sec_per_batch = duration / 20.0
          examples_per_sec = FLAGS.batch_size / sec_per_batch
          print('%s: [%d batches out of %d] (%.1f examples/sec; %.3f'
                'sec/batch)' % (datetime.now(), step, num_iter,
                                examples_per_sec, sec_per_batch))
          start_time = time.time()

      # Compute precision @ 1.
      precision_at_1 = count_top_1 / total_sample_count
      recall_at_5 = count_top_5 / total_sample_count
      print('%s: precision @ 1 = %.4f recall @ 5 = %.4f [%d examples]' %
            (datetime.now(), precision_at_1, recall_at_5, total_sample_count))

      summary = tf.Summary()
      summary.ParseFromString(sess.run(summary_op))
      summary.value.add(tag='Precision @ 1', simple_value=precision_at_1)
      summary.value.add(tag='Recall @ 5', simple_value=recall_at_5)
      summary_writer.add_summary(summary, global_step)

    except Exception as e:  # pylint: disable=broad-except
      coord.request_stop(e)

    coord.request_stop()
    coord.join(threads, stop_grace_period_secs=10)
    # evaluate(FLAGS.subset)


def evaluate(dataset):
  """Evaluate model on Dataset for a number of steps."""
  with tf.Graph().as_default():
    # Get images and labels from the dataset.

    # with tf.Session() as sess:
    #   images, labels = image_processing.inputs(dataset)
    #   print ("lable")
    #   sess.run (labels.eval())
    images, labels, filenames = image_processing.inputs(dataset)

    # Number of classes in the Dataset label set plus 1.
    # Label 0 is reserved for an (unused) background class.
    num_classes = dataset.num_classes() + 1

    # Build a Graph that computes the logits predictions from the
    # inference model.
    logits, _ = inception.inference(images, num_classes)
    print ("logits")
    # print (tf.cast(logits,tf.float32).eval())
    # print ("_")
    # print (tf.cast(_,tf.float32).eval())

    # Calculate predictions.
    top_1_op = tf.nn.in_top_k(logits, labels, 1)
    top_5_op = tf.nn.in_top_k(logits, labels, 5)

    # Restore the moving average version of the learned variables for eval.
    variable_averages = tf.train.ExponentialMovingAverage(
        inception.MOVING_AVERAGE_DECAY)
    variables_to_restore = variable_averages.variables_to_restore()
    saver = tf.train.Saver(variables_to_restore)

    # Build the summary operation based on the TF collection of Summaries.
    summary_op = tf.summary.merge_all()

    graph_def = tf.get_default_graph().as_graph_def()
    summary_writer = tf.summary.FileWriter(FLAGS.eval_dir,
                                            graph_def=graph_def)

    while True:

      _eval_once(saver, summary_writer, filenames, logits, labels, top_1_op, top_5_op, summary_op)
      if FLAGS.run_once:
        break
      time.sleep(FLAGS.eval_interval_secs)
    

    # sess = tf.InteractiveSession()
    # print("label")
    # # label_ = sess.run([labels])
    # print(labels.eval())
    # sess.close()

image_procession.py

# Copyright 2016 Google Inc. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Read and preprocess image data.

 Image processing occurs on a single image at a time. Image are read and
 preprocessed in parallel across multiple threads. The resulting images
 are concatenated together to form a single batch for training or evaluation.

 -- Provide processed image data for a network:
 inputs: Construct batches of evaluation examples of images.
 distorted_inputs: Construct batches of training examples of images.
 batch_inputs: Construct batches of training or evaluation examples of images.

 -- Data processing:
 parse_example_proto: Parses an Example proto containing a training example
   of an image.

 -- Image decoding:
 decode_jpeg: Decode a JPEG encoded string into a 3-D float32 Tensor.

 -- Image preprocessing:
 image_preprocessing: Decode and preprocess one image for evaluation or training
 distort_image: Distort one image for training a network.
 eval_image: Prepare one image for evaluation.
 distort_color: Distort the color in one image for training.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import tensorflow as tf

FLAGS = tf.app.flags.FLAGS

tf.app.flags.DEFINE_integer('batch_size', 32,
                            """Number of images to process in a batch.""")
tf.app.flags.DEFINE_integer('image_size', 299,
                            """Provide square images of this size.""")
tf.app.flags.DEFINE_integer('num_preprocess_threads', 4,
                            """Number of preprocessing threads per tower. """
                            """Please make this a multiple of 4.""")
tf.app.flags.DEFINE_integer('num_readers', 4,
                            """Number of parallel readers during train.""")

# Images are preprocessed asynchronously using multiple threads specified by
# --num_preprocss_threads and the resulting processed images are stored in a
# random shuffling queue. The shuffling queue dequeues --batch_size images
# for processing on a given Inception tower. A larger shuffling queue guarantees
# better mixing across examples within a batch and results in slightly higher
# predictive performance in a trained model. Empirically,
# --input_queue_memory_factor=16 works well. A value of 16 implies a queue size
# of 1024*16 images. Assuming RGB 299x299 images, this implies a queue size of
# 16GB. If the machine is memory limited, then decrease this factor to
# decrease the CPU memory footprint, accordingly.
tf.app.flags.DEFINE_integer('input_queue_memory_factor', 16,
                            """Size of the queue of preprocessed images. """
                            """Default is ideal but try smaller values, e.g. """
                            """4, 2 or 1, if host memory is constrained. See """
                            """comments in code for more details.""")


# def inputs(dataset, batch_size=None, num_preprocess_threads=None):
#   """Generate batches of ImageNet images for evaluation.
#
#   Use this function as the inputs for evaluating a network.
#
#   Note that some (minimal) image preprocessing occurs during evaluation
#   including central cropping and resizing of the image to fit the network.
#
#   Args:
#     dataset: instance of Dataset class specifying the dataset.
#     batch_size: integer, number of examples in batch
#     num_preprocess_threads: integer, total number of preprocessing threads but
#       None defaults to FLAGS.num_preprocess_threads.
#
#   Returns:
#     images: Images. 4D tensor of size [batch_size, FLAGS.image_size,
#                                        image_size, 3].
#     labels: 1-D integer Tensor of [FLAGS.batch_size].
#   """
#   if not batch_size:
#     batch_size = FLAGS.batch_size
#
#   # Force all input processing onto CPU in order to reserve the GPU for
#   # the forward inference and back-propagation.
#   with tf.device('/cpu:0'):
#     images, labels = batch_inputs(
#         dataset, batch_size, train=False,
#         num_preprocess_threads=num_preprocess_threads,
#         num_readers=1)
#
#   return images, labels
#
#
# def distorted_inputs(dataset, batch_size=None, num_preprocess_threads=None):
#   """Generate batches of distorted versions of ImageNet images.
#
#   Use this function as the inputs for training a network.
#
#   Distorting images provides a useful technique for augmenting the data
#   set during training in order to make the network invariant to aspects
#   of the image that do not effect the label.
#
#   Args:
#     dataset: instance of Dataset class specifying the dataset.
#     batch_size: integer, number of examples in batch
#     num_preprocess_threads: integer, total number of preprocessing threads but
#       None defaults to FLAGS.num_preprocess_threads.
#
#   Returns:
#     images: Images. 4D tensor of size [batch_size, FLAGS.image_size,
#                                        FLAGS.image_size, 3].
#     labels: 1-D integer Tensor of [batch_size].
#   """
#   if not batch_size:
#     batch_size = FLAGS.batch_size
#
#   # Force all input processing onto CPU in order to reserve the GPU for
#   # the forward inference and back-propagation.
#   with tf.device('/cpu:0'):
#     images, labels = batch_inputs(
#         dataset, batch_size, train=True,
#         num_preprocess_threads=num_preprocess_threads,
#         num_readers=FLAGS.num_readers)
#   return images, labels
#
#
# def decode_jpeg(image_buffer, scope=None):
#   """Decode a JPEG string into one 3-D float image Tensor.
#
#   Args:
#     image_buffer: scalar string Tensor.
#     scope: Optional scope for name_scope.
#   Returns:
#     3-D float Tensor with values ranging from [0, 1).
#   """
#   with tf.name_scope(values=[image_buffer], name=scope,
#                      default_name='decode_jpeg'):
#     # Decode the string as an RGB JPEG.
#     # Note that the resulting image contains an unknown height and width
#     # that is set dynamically by decode_jpeg. In other words, the height
#     # and width of image is unknown at compile-time.
#     image = tf.image.decode_jpeg(image_buffer, channels=3)
#
#     # After this point, all image pixels reside in [0,1)
#     # until the very end, when they're rescaled to (-1, 1).  The various
#     # adjust_* ops all require this range for dtype float.
#     image = tf.image.convert_image_dtype(image, dtype=tf.float32)
#     return image
#
#
# def distort_color(image, thread_id=0, scope=None):
#   """Distort the color of the image.
#
#   Each color distortion is non-commutative and thus ordering of the color ops
#   matters. Ideally we would randomly permute the ordering of the color ops.
#   Rather then adding that level of complication, we select a distinct ordering
#   of color ops for each preprocessing thread.
#
#   Args:
#     image: Tensor containing single image.
#     thread_id: preprocessing thread ID.
#     scope: Optional scope for name_scope.
#   Returns:
#     color-distorted image
#   """
#   with tf.name_scope(values=[image], name=scope, default_name='distort_color'):
#     color_ordering = thread_id % 2
#
#     if color_ordering == 0:
#       image = tf.image.random_brightness(image, max_delta=32. / 255.)
#       image = tf.image.random_saturation(image, lower=0.5, upper=1.5)
#       image = tf.image.random_hue(image, max_delta=0.2)
#       image = tf.image.random_contrast(image, lower=0.5, upper=1.5)
#     elif color_ordering == 1:
#       image = tf.image.random_brightness(image, max_delta=32. / 255.)
#       image = tf.image.random_contrast(image, lower=0.5, upper=1.5)
#       image = tf.image.random_saturation(image, lower=0.5, upper=1.5)
#       image = tf.image.random_hue(image, max_delta=0.2)
#
#     # The random_* ops do not necessarily clamp.
#     image = tf.clip_by_value(image, 0.0, 1.0)
#     return image
#
#
# def distort_image(image, height, width, bbox, thread_id=0, scope=None):
#   """Distort one image for training a network.
#
#   Distorting images provides a useful technique for augmenting the data
#   set during training in order to make the network invariant to aspects
#   of the image that do not effect the label.
#
#   Args:
#     image: 3-D float Tensor of image
#     height: integer
#     width: integer
#     bbox: 3-D float Tensor of bounding boxes arranged [1, num_boxes, coords]
#       where each coordinate is [0, 1) and the coordinates are arranged
#       as [ymin, xmin, ymax, xmax].
#     thread_id: integer indicating the preprocessing thread.
#     scope: Optional scope for name_scope.
#   Returns:
#     3-D float Tensor of distorted image used for training.
#   """
#   with tf.name_scope(values=[image, height, width, bbox], name=scope,
#                      default_name='distort_image'):
#     # Each bounding box has shape [1, num_boxes, box coords] and
#     # the coordinates are ordered [ymin, xmin, ymax, xmax].
#
#     # Display the bounding box in the first thread only.
#     if not thread_id:
#       image_with_box = tf.image.draw_bounding_boxes(tf.expand_dims(image, 0),
#                                                     bbox)
#       tf.summary.image('image_with_bounding_boxes', image_with_box)
#
#   # A large fraction of image datasets contain a human-annotated bounding
#   # box delineating the region of the image containing the object of interest.
#   # We choose to create a new bounding box for the object which is a randomly
#   # distorted version of the human-annotated bounding box that obeys an allowed
#   # range of aspect ratios, sizes and overlap with the human-annotated
#   # bounding box. If no box is supplied, then we assume the bounding box is
#   # the entire image.
#     sample_distorted_bounding_box = tf.image.sample_distorted_bounding_box(
#         tf.shape(image),
#         bounding_boxes=bbox,
#         min_object_covered=0.1,
#         aspect_ratio_range=[0.75, 1.33],
#         area_range=[0.05, 1.0],
#         max_attempts=100,
#         use_image_if_no_bounding_boxes=True)
#     bbox_begin, bbox_size, distort_bbox = sample_distorted_bounding_box
#     if not thread_id:
#       image_with_distorted_box = tf.image.draw_bounding_boxes(
#           tf.expand_dims(image, 0), distort_bbox)
#       tf.summary.image('images_with_distorted_bounding_box',
#                        image_with_distorted_box)
#
#     # Crop the image to the specified bounding box.
#     distorted_image = tf.slice(image, bbox_begin, bbox_size)
#
#     # This resizing operation may distort the images because the aspect
#     # ratio is not respected. We select a resize method in a round robin
#     # fashion based on the thread number.
#     # Note that ResizeMethod contains 4 enumerated resizing methods.
#     resize_method = thread_id % 4
#     distorted_image = tf.image.resize_images(distorted_image, [height, width],
#                                              method=resize_method)
#     # Restore the shape since the dynamic slice based upon the bbox_size loses
#     # the third dimension.
#     distorted_image.set_shape([height, width, 3])
#     if not thread_id:
#       tf.summary.image('cropped_resized_image',
#                        tf.expand_dims(distorted_image, 0))
#
#     # Randomly flip the image horizontally.
#     distorted_image = tf.image.random_flip_left_right(distorted_image)
#
#     # Randomly distort the colors.
#     distorted_image = distort_color(distorted_image, thread_id)
#
#     if not thread_id:
#       tf.summary.image('final_distorted_image',
#                        tf.expand_dims(distorted_image, 0))
#     return distorted_image
#
#
# def eval_image(image, height, width, scope=None):
#   """Prepare one image for evaluation.
#
#   Args:
#     image: 3-D float Tensor
#     height: integer
#     width: integer
#     scope: Optional scope for name_scope.
#   Returns:
#     3-D float Tensor of prepared image.
#   """
#   with tf.name_scope(values=[image, height, width], name=scope,
#                      default_name='eval_image'):
#     # Crop the central region of the image with an area containing 87.5% of
#     # the original image.
#     image = tf.image.central_crop(image, central_fraction=0.875)
#
#     # Resize the image to the original height and width.
#     image = tf.expand_dims(image, 0)
#     image = tf.image.resize_bilinear(image, [height, width],
#                                      align_corners=False)
#     image = tf.squeeze(image, [0])
#     return image
#
#
# def image_preprocessing(image_buffer, bbox, train, thread_id=0):
#   """Decode and preprocess one image for evaluation or training.
#
#   Args:
#     image_buffer: JPEG encoded string Tensor
#     bbox: 3-D float Tensor of bounding boxes arranged [1, num_boxes, coords]
#       where each coordinate is [0, 1) and the coordinates are arranged as
#       [ymin, xmin, ymax, xmax].
#     train: boolean
#     thread_id: integer indicating preprocessing thread
#
#   Returns:
#     3-D float Tensor containing an appropriately scaled image
#
#   Raises:
#     ValueError: if user does not provide bounding box
#   """
#   if bbox is None:
#     raise ValueError('Please supply a bounding box.')
#
#   image = decode_jpeg(image_buffer)
#   height = FLAGS.image_size
#   width = FLAGS.image_size
#
#   if train:
#     image = distort_image(image, height, width, bbox, thread_id)
#   else:
#     image = eval_image(image, height, width)
#
#   # Finally, rescale to [-1,1] instead of [0, 1)
#   image = tf.subtract(image, 0.5)
#   image = tf.multiply(image, 2.0)
#   return image
#
#
# def parse_example_proto(example_serialized):
#   """Parses an Example proto containing a training example of an image.
#
#   The output of the build_image_data.py image preprocessing script is a dataset
#   containing serialized Example protocol buffers. Each Example proto contains
#   the following fields:
#
#     image/height: 462
#     image/width: 581
#     image/colorspace: 'RGB'
#     image/channels: 3
#     image/class/label: 615
#     image/class/synset: 'n03623198'
#     image/class/text: 'knee pad'
#     image/object/bbox/xmin: 0.1
#     image/object/bbox/xmax: 0.9
#     image/object/bbox/ymin: 0.2
#     image/object/bbox/ymax: 0.6
#     image/object/bbox/label: 615
#     image/format: 'JPEG'
#     image/filename: 'ILSVRC2012_val_00041207.JPEG'
#     image/encoded: <JPEG encoded string>
#
#   Args:
#     example_serialized: scalar Tensor tf.string containing a serialized
#       Example protocol buffer.
#
#   Returns:
#     image_buffer: Tensor tf.string containing the contents of a JPEG file.
#     label: Tensor tf.int32 containing the label.
#     bbox: 3-D float Tensor of bounding boxes arranged [1, num_boxes, coords]
#       where each coordinate is [0, 1) and the coordinates are arranged as
#       [ymin, xmin, ymax, xmax].
#     text: Tensor tf.string containing the human-readable label.
#   """
#   # Dense features in Example proto.
#   feature_map = {
#       'image/encoded': tf.FixedLenFeature([], dtype=tf.string,
#                                           default_value=''),
#       'image/class/label': tf.FixedLenFeature([1], dtype=tf.int64,
#                                               default_value=-1),
#       'image/class/text': tf.FixedLenFeature([], dtype=tf.string,
#                                              default_value=''),
#   }
#   sparse_float32 = tf.VarLenFeature(dtype=tf.float32)
#   # Sparse features in Example proto.
#   feature_map.update(
#       {k: sparse_float32 for k in ['image/object/bbox/xmin',
#                                    'image/object/bbox/ymin',
#                                    'image/object/bbox/xmax',
#                                    'image/object/bbox/ymax']})
#
#   features = tf.parse_single_example(example_serialized, feature_map)
#   label = tf.cast(features['image/class/label'], dtype=tf.int32)
#
#   xmin = tf.expand_dims(features['image/object/bbox/xmin'].values, 0)
#   ymin = tf.expand_dims(features['image/object/bbox/ymin'].values, 0)
#   xmax = tf.expand_dims(features['image/object/bbox/xmax'].values, 0)
#   ymax = tf.expand_dims(features['image/object/bbox/ymax'].values, 0)
#
#   # Note that we impose an ordering of (y, x) just to make life difficult.
#   bbox = tf.concat(axis=0, values=[ymin, xmin, ymax, xmax])
#
#   # Force the variable number of bounding boxes into the shape
#   # [1, num_boxes, coords].
#   bbox = tf.expand_dims(bbox, 0)
#   bbox = tf.transpose(bbox, [0, 2, 1])
#
#   return features['image/encoded'], label, bbox, features['image/class/text']
#
#
# def batch_inputs(dataset, batch_size, train, num_preprocess_threads=None,
#                  num_readers=1):
#   """Contruct batches of training or evaluation examples from the image dataset.
#
#   Args:
#     dataset: instance of Dataset class specifying the dataset.
#       See dataset.py for details.
#     batch_size: integer
#     train: boolean
#     num_preprocess_threads: integer, total number of preprocessing threads
#     num_readers: integer, number of parallel readers
#
#   Returns:
#     images: 4-D float Tensor of a batch of images
#     labels: 1-D integer Tensor of [batch_size].
#
#   Raises:
#     ValueError: if data is not found
#   """
#   with tf.name_scope('batch_processing'):
#     data_files = dataset.data_files()
#     if data_files is None:
#       raise ValueError('No data files found for this dataset')
#
#     # Create filename_queue
#     if train:
#       filename_queue = tf.train.string_input_producer(data_files,
#                                                       shuffle=True,
#                                                       capacity=16)
#     else:
#       filename_queue = tf.train.string_input_producer(data_files,
#                                                       shuffle=False,
#                                                       capacity=1)
#     if num_preprocess_threads is None:
#       num_preprocess_threads = FLAGS.num_preprocess_threads
#
#     if num_preprocess_threads % 4:
#       raise ValueError('Please make num_preprocess_threads a multiple '
#                        'of 4 (%d % 4 != 0).', num_preprocess_threads)
#
#     if num_readers is None:
#       num_readers = FLAGS.num_readers
#
#     if num_readers < 1:
#       raise ValueError('Please make num_readers at least 1')
#
#     # Approximate number of examples per shard.
#     examples_per_shard = 1024
#     # Size the random shuffle queue to balance between good global
#     # mixing (more examples) and memory use (fewer examples).
#     # 1 image uses 299*299*3*4 bytes = 1MB
#     # The default input_queue_memory_factor is 16 implying a shuffling queue
#     # size: examples_per_shard * 16 * 1MB = 17.6GB
#     min_queue_examples = examples_per_shard * FLAGS.input_queue_memory_factor
#     if train:
#       examples_queue = tf.RandomShuffleQueue(
#           capacity=min_queue_examples + 3 * batch_size,
#           min_after_dequeue=min_queue_examples,
#           dtypes=[tf.string])
#     else:
#       examples_queue = tf.FIFOQueue(
#           capacity=examples_per_shard + 3 * batch_size,
#           dtypes=[tf.string])
#
#     # Create multiple readers to populate the queue of examples.
#     if num_readers > 1:
#       enqueue_ops = []
#       for _ in range(num_readers):
#         reader = dataset.reader()
#         _, value = reader.read(filename_queue)
#         enqueue_ops.append(examples_queue.enqueue([value]))
#
#       tf.train.queue_runner.add_queue_runner(
#           tf.train.queue_runner.QueueRunner(examples_queue, enqueue_ops))
#       example_serialized = examples_queue.dequeue()
#     else:
#       reader = dataset.reader()
#       _, example_serialized = reader.read(filename_queue)
#
#     images_and_labels = []
#     for thread_id in range(num_preprocess_threads):
#       # Parse a serialized Example proto to extract the image and metadata.
#       image_buffer, label_index, bbox, _ = parse_example_proto(
#           example_serialized)
#       image = image_preprocessing(image_buffer, bbox, train, thread_id)
#       images_and_labels.append([image, label_index])
#
#     images, label_index_batch = tf.train.batch_join(
#         images_and_labels,
#         batch_size=batch_size,
#         capacity=2 * num_preprocess_threads * batch_size)
#
#     # Reshape images into these desired dimensions.
#     height = FLAGS.image_size
#     width = FLAGS.image_size
#     depth = 3
#
#     images = tf.cast(images, tf.float32)
#     images = tf.reshape(images, shape=[batch_size, height, width, depth])
#
#     # Display the training images in the visualizer.
#     tf.summary.image('images', images)
#
#     return images, tf.reshape(label_index_batch, [batch_size])
def inputs(dataset, batch_size=None, num_preprocess_threads=None):
  """Generate batches of ImageNet images for evaluation.

  Use this function as the inputs for evaluating a network.

  Note that some (minimal) image preprocessing occurs during evaluation
  including central cropping and resizing of the image to fit the network.

  Args:
    dataset: instance of Dataset class specifying the dataset.
    batch_size: integer, number of examples in batch
    num_preprocess_threads: integer, total number of preprocessing threads but
      None defaults to FLAGS.num_preprocess_threads.

  Returns:
    images: Images. 4D tensor of size [batch_size, FLAGS.image_size,
                                       image_size, 3].
    labels: 1-D integer Tensor of [FLAGS.batch_size].
  """
  if not batch_size:
    batch_size = FLAGS.batch_size

  # Force all input processing onto CPU in order to reserve the GPU for
  # the forward inference and back-propagation.
  with tf.device('/cpu:0'):
    images, labels, filenames = batch_inputs(
        dataset, batch_size, train=False,
        num_preprocess_threads=num_preprocess_threads,
        num_readers=1)

  return images, labels, filenames


def distorted_inputs(dataset, batch_size=None, num_preprocess_threads=None):
  """Generate batches of distorted versions of ImageNet images.

  Use this function as the inputs for training a network.

  Distorting images provides a useful technique for augmenting the data
  set during training in order to make the network invariant to aspects
  of the image that do not effect the label.

  Args:
    dataset: instance of Dataset class specifying the dataset.
    batch_size: integer, number of examples in batch
    num_preprocess_threads: integer, total number of preprocessing threads but
      None defaults to FLAGS.num_preprocess_threads.

  Returns:
    images: Images. 4D tensor of size [batch_size, FLAGS.image_size,
                                       FLAGS.image_size, 3].
    labels: 1-D integer Tensor of [batch_size].
  """
  if not batch_size:
    batch_size = FLAGS.batch_size

  # Force all input processing onto CPU in order to reserve the GPU for
  # the forward inference and back-propagation.
  with tf.device('/cpu:0'):
    images, labels, _ = batch_inputs(
        dataset, batch_size, train=True,
        num_preprocess_threads=num_preprocess_threads,
        num_readers=FLAGS.num_readers)
  return images, labels


def decode_jpeg(image_buffer, scope=None):
  """Decode a JPEG string into one 3-D float image Tensor.

  Args:
    image_buffer: scalar string Tensor.
    scope: Optional scope for op_scope.
  Returns:
    3-D float Tensor with values ranging from [0, 1).
  """
  with tf.op_scope([image_buffer], scope, 'decode_jpeg'):
    # Decode the string as an RGB JPEG.
    # Note that the resulting image contains an unknown height and width
    # that is set dynamically by decode_jpeg. In other words, the height
    # and width of image is unknown at compile-time.
    image = tf.image.decode_jpeg(image_buffer, channels=3)

    # After this point, all image pixels reside in [0,1)
    # until the very end, when they're rescaled to (-1, 1).  The various
    # adjust_* ops all require this range for dtype float.
    image = tf.image.convert_image_dtype(image, dtype=tf.float32)
    return image


def distort_color(image, thread_id=0, scope=None):
  """Distort the color of the image.

  Each color distortion is non-commutative and thus ordering of the color ops
  matters. Ideally we would randomly permute the ordering of the color ops.
  Rather then adding that level of complication, we select a distinct ordering
  of color ops for each preprocessing thread.

  Args:
    image: Tensor containing single image.
    thread_id: preprocessing thread ID.
    scope: Optional scope for op_scope.
  Returns:
    color-distorted image
  """
  with tf.op_scope([image], scope, 'distort_color'):
    color_ordering = thread_id % 2

    if color_ordering == 0:
      image = tf.image.random_brightness(image, max_delta=32. / 255.)
      image = tf.image.random_saturation(image, lower=0.5, upper=1.5)
      image = tf.image.random_hue(image, max_delta=0.2)
      image = tf.image.random_contrast(image, lower=0.5, upper=1.5)
    elif color_ordering == 1:
      image = tf.image.random_brightness(image, max_delta=32. / 255.)
      image = tf.image.random_contrast(image, lower=0.5, upper=1.5)
      image = tf.image.random_saturation(image, lower=0.5, upper=1.5)
      image = tf.image.random_hue(image, max_delta=0.2)

    # The random_* ops do not necessarily clamp.
    image = tf.clip_by_value(image, 0.0, 1.0)
    return image


def distort_image(image, height, width, bbox, thread_id=0, scope=None):
  """Distort one image for training a network.

  Distorting images provides a useful technique for augmenting the data
  set during training in order to make the network invariant to aspects
  of the image that do not effect the label.

  Args:
    image: 3-D float Tensor of image
    height: integer
    width: integer
    bbox: 3-D float Tensor of bounding boxes arranged [1, num_boxes, coords]
      where each coordinate is [0, 1) and the coordinates are arranged
      as [ymin, xmin, ymax, xmax].
    thread_id: integer indicating the preprocessing thread.
    scope: Optional scope for op_scope.
  Returns:
    3-D float Tensor of distorted image used for training.
  """
  with tf.op_scope([image, height, width, bbox], scope, 'distort_image'):
    # Each bounding box has shape [1, num_boxes, box coords] and
    # the coordinates are ordered [ymin, xmin, ymax, xmax].

    # Display the bounding box in the first thread only.
    if not thread_id:
      image_with_box = tf.image.draw_bounding_boxes(tf.expand_dims(image, 0),
                                                    bbox)
      tf.image_summary('image_with_bounding_boxes', image_with_box)

  # A large fraction of image datasets contain a human-annotated bounding
  # box delineating the region of the image containing the object of interest.
  # We choose to create a new bounding box for the object which is a randomly
  # distorted version of the human-annotated bounding box that obeys an allowed
  # range of aspect ratios, sizes and overlap with the human-annotated
  # bounding box. If no box is supplied, then we assume the bounding box is
  # the entire image.
    sample_distorted_bounding_box = tf.image.sample_distorted_bounding_box(
        tf.shape(image),
        bounding_boxes=bbox,
        min_object_covered=0.1,
        aspect_ratio_range=[0.75, 1.33],
        area_range=[0.05, 1.0],
        max_attempts=100,
        use_image_if_no_bounding_boxes=True)
    bbox_begin, bbox_size, distort_bbox = sample_distorted_bounding_box
    if not thread_id:
      image_with_distorted_box = tf.image.draw_bounding_boxes(
          tf.expand_dims(image, 0), distort_bbox)
      tf.image_summary('images_with_distorted_bounding_box',
                       image_with_distorted_box)

    # Crop the image to the specified bounding box.
    distorted_image = tf.slice(image, bbox_begin, bbox_size)

    # This resizing operation may distort the images because the aspect
    # ratio is not respected. We select a resize method in a round robin
    # fashion based on the thread number.
    # Note that ResizeMethod contains 4 enumerated resizing methods.
    resize_method = thread_id % 4
    distorted_image = tf.image.resize_images(distorted_image, [height, width],
                                             method=resize_method)
    # Restore the shape since the dynamic slice based upon the bbox_size loses
    # the third dimension.
    distorted_image.set_shape([height, width, 3])
    if not thread_id:
      tf.image_summary('cropped_resized_image',
                       tf.expand_dims(distorted_image, 0))

    # Randomly flip the image horizontally.
    distorted_image = tf.image.random_flip_left_right(distorted_image)

    # Randomly distort the colors.
    distorted_image = distort_color(distorted_image, thread_id)

    if not thread_id:
      tf.image_summary('final_distorted_image',
                       tf.expand_dims(distorted_image, 0))
    return distorted_image


def eval_image(image, height, width, scope=None):
  """Prepare one image for evaluation.

  Args:
    image: 3-D float Tensor
    height: integer
    width: integer
    scope: Optional scope for op_scope.
  Returns:
    3-D float Tensor of prepared image.
  """
  with tf.op_scope([image, height, width], scope, 'eval_image'):
    # Crop the central region of the image with an area containing 87.5% of
    # the original image.
    image = tf.image.central_crop(image, central_fraction=0.875)

    # Resize the image to the original height and width.
    image = tf.expand_dims(image, 0)
    image = tf.image.resize_bilinear(image, [height, width],
                                     align_corners=False)
    image = tf.squeeze(image, [0])
    return image


def image_preprocessing(image_buffer, bbox, train, thread_id=0):
  """Decode and preprocess one image for evaluation or training.

  Args:
    image_buffer: JPEG encoded string Tensor
    bbox: 3-D float Tensor of bounding boxes arranged [1, num_boxes, coords]
      where each coordinate is [0, 1) and the coordinates are arranged as
      [ymin, xmin, ymax, xmax].
    train: boolean
    thread_id: integer indicating preprocessing thread

  Returns:
    3-D float Tensor containing an appropriately scaled image

  Raises:
    ValueError: if user does not provide bounding box
  """
  if bbox is None:
    raise ValueError('Please supply a bounding box.')

  image = decode_jpeg(image_buffer)
  height = FLAGS.image_size
  width = FLAGS.image_size

  if train:
    image = distort_image(image, height, width, bbox, thread_id)
  else:
    image = eval_image(image, height, width)

  # Finally, rescale to [-1,1] instead of [0, 1)
  image = tf.subtract(image, 0.5)
  image = tf.multiply(image, 2.0)
  return image

def debug_print(y):

    with tf.Session():
     print(y.eval())

def parse_example_proto(example_serialized):
  """Parses an Example proto containing a training example of an image.

  The output of the build_image_data.py image preprocessing script is a dataset
  containing serialized Example protocol buffers. Each Example proto contains
  the following fields:

    image/height: 462
    image/width: 581
    image/colorspace: 'RGB'
    image/channels: 3
    image/class/label: 615
    image/class/synset: 'n03623198'
    image/class/text: 'knee pad'
    image/object/bbox/xmin: 0.1
    image/object/bbox/xmax: 0.9
    image/object/bbox/ymin: 0.2
    image/object/bbox/ymax: 0.6
    image/object/bbox/label: 615
    image/format: 'JPEG'
    image/filename: 'ILSVRC2012_val_00041207.JPEG'
    image/encoded: <JPEG encoded string>

  Args:
    example_serialized: scalar Tensor tf.string containing a serialized
      Example protocol buffer.

  Returns:
    image_buffer: Tensor tf.string containing the contents of a JPEG file.
    label: Tensor tf.int32 containing the label.
    bbox: 3-D float Tensor of bounding boxes arranged [1, num_boxes, coords]
      where each coordinate is [0, 1) and the coordinates are arranged as
      [ymin, xmin, ymax, xmax].
    text: Tensor tf.string containing the human-readable label.
  """
  # Dense features in Example proto.
  feature_map = {
      'image/encoded': tf.FixedLenFeature([], dtype=tf.string,
                                          default_value=''),
      'image/class/label': tf.FixedLenFeature([1], dtype=tf.int64,
                                              default_value=-1),
      'image/class/text': tf.FixedLenFeature([], dtype=tf.string,
                                             default_value=''),
      'image/filename': tf.FixedLenFeature([], dtype=tf.string,
                                             default_value=''),
  }
  sparse_float32 = tf.VarLenFeature(dtype=tf.float32)
  # Sparse features in Example proto.
  feature_map.update(
      {k: sparse_float32 for k in ['image/object/bbox/xmin',
                                   'image/object/bbox/ymin',
                                   'image/object/bbox/xmax',
                                   'image/object/bbox/ymax']})

  features = tf.parse_single_example(example_serialized, feature_map)
  label = tf.cast(features['image/class/label'], dtype=tf.int32)

  xmin = tf.expand_dims(features['image/object/bbox/xmin'].values, 0)
  ymin = tf.expand_dims(features['image/object/bbox/ymin'].values, 0)
  xmax = tf.expand_dims(features['image/object/bbox/xmax'].values, 0)
  ymax = tf.expand_dims(features['image/object/bbox/ymax'].values, 0)

  # Note that we impose an ordering of (y, x) just to make life difficult.
  bbox = tf.concat(axis=0, values=[ymin, xmin, ymax, xmax])

  # Force the variable number of bounding boxes into the shape
  # [1, num_boxes, coords].
  bbox = tf.expand_dims(bbox, 0)
  bbox = tf.transpose(bbox, [0, 2, 1])

  return features['image/encoded'], label, bbox, features['image/class/text'], features['image/filename']


def batch_inputs(dataset, batch_size, train, num_preprocess_threads=None,
                 num_readers=1):
  """Contruct batches of training or evaluation examples from the image dataset.

  Args:
    dataset: instance of Dataset class specifying the dataset.
      See dataset.py for details.
    batch_size: integer
    train: boolean
    num_preprocess_threads: integer, total number of preprocessing threads
    num_readers: integer, number of parallel readers

  Returns:
    images: 4-D float Tensor of a batch of images
    labels: 1-D integer Tensor of [batch_size].
    filename list: the list of filename

  Raises:
    ValueError: if data is not found
  """
  with tf.name_scope('batch_processing'):
    data_files = dataset.data_files()
    if data_files is None:
      raise ValueError('No data files found for this dataset')

    # Create filename_queue
    if train:
      filename_queue = tf.train.string_input_producer(data_files,
                                                      shuffle=True,
                                                      capacity=16)
    else:
      filename_queue = tf.train.string_input_producer(data_files,
                                                      shuffle=False,
                                                        capacity=1)
    if num_preprocess_threads is None:
      num_preprocess_threads = FLAGS.num_preprocess_threads

    if num_preprocess_threads % 4:
      _=1
      #raise ValueError('Please make num_preprocess_threads a multiple '
      #                 'of 4 (%d % 4 != 0).', num_preprocess_threads)

    if num_readers is None:
      num_readers = FLAGS.num_readers

    if num_readers < 1:
      raise ValueError('Please make num_readers at least 1')

    # Approximate number of examples per shard.
    examples_per_shard = 1024
    # Size the random shuffle queue to balance between good global
    # mixing (more examples) and memory use (fewer examples).
    # 1 image uses 299*299*3*4 bytes = 1MB
    # The default input_queue_memory_factor is 16 implying a shuffling queue
    # size: examples_per_shard * 16 * 1MB = 17.6GB
    min_queue_examples = examples_per_shard * FLAGS.input_queue_memory_factor
    if train:
      examples_queue = tf.RandomShuffleQueue(
          capacity=min_queue_examples + 3 * batch_size,
          min_after_dequeue=min_queue_examples,
          dtypes=[tf.string])
    else:
      examples_queue = tf.FIFOQueue(
          capacity=examples_per_shard + 3 * batch_size,
          dtypes=[tf.string])

    # Create multiple readers to populate the queue of examples.
    if num_readers > 1:
      enqueue_ops = []
      for _ in range(num_readers):
        reader = dataset.reader()
        _, value = reader.read(filename_queue)
        enqueue_ops.append(examples_queue.enqueue([value]))

      tf.train.queue_runner.add_queue_runner(
          tf.train.queue_runner.QueueRunner(examples_queue, enqueue_ops))
      example_serialized = examples_queue.dequeue()
    else:
      reader = dataset.reader()
      _, example_serialized = reader.read(filename_queue)

    images_and_labels = []
    for thread_id in range(num_preprocess_threads):
      # Parse a serialized Example proto to extract the image and metadata.
      image_buffer, label_index, bbox, _, filename = parse_example_proto(
          example_serialized)
      image = image_preprocessing(image_buffer, bbox, train, thread_id)
      images_and_labels.append([image, label_index,filename])

    images, label_index_batch,filenames = tf.train.batch_join(
        images_and_labels,
        batch_size=batch_size,
        capacity=3 * num_preprocess_threads * batch_size)

    # Reshape images into these desired dimensions.
    height = FLAGS.image_size
    width = FLAGS.image_size
    depth = 3

    images = tf.cast(images, tf.float32)
    images = tf.reshape(images, shape=[batch_size, height, width, depth])

    # Display the training images in the visualizer.
    tf.summary.image('images', images)

    return images, tf.reshape(label_index_batch, [batch_size]), tf.reshape(filenames, [batch_size])

这些大概就是这段时间遇到问题的总结  写下来虽然不是很多  但是作为小白 确实花了很多时间去折腾