第三十四节，目标检测之谷歌Object Detection API源码解析

我们在第三十二节，使用谷歌Object Detection API进行目标检测、训练新的模型(使用VOC 2012数据集)那一节我们介绍了如何使用谷歌Object Detection API进行目标检测，以及如何使用谷歌提供的目标检测模型训练自己的数据。在训练自己的数据集时，主要包括以下几步：

• 制作自己的数据集，注意这里数据集在进行标注时，需要按照一定的格式。然后调object_detection\dataset_tools下对应的脚本生成tfrecord文件。如下图，如果我们想调用create_pascal_tf_record.py文件生成tfrecord文件，那么我们的数据集要和voc 2012数据集的标注方式一样。你也可以通过解读create_pascal_tf_record.py文件了解我们的数据集的标注方式。

• 下载我们所要使用的目标检测模型，进行预训练，不然从头开始训练时间成本会很高。
• 在object_detection/samples/configs文件夹下有一些配置文件，选择与我们所要使用的目标检测模型相对应的配置文件，并进行一些修改。
• 使用object_detection/train.py文件进行训练。
• 使用export_inference_graph.py脚本导出训练好的模型，并进行目标检测。

在这里我主要解析一下train.py文件的工作流程。

一 train.py文件解析

先附上源码：

# Copyright 2017 The TensorFlow Authors. 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.
# ==============================================================================

r"""Training executable for detection models.

This executable is used to train DetectionModels. There are two ways of
configuring the training job:

1) A single pipeline_pb2.TrainEvalPipelineConfig configuration file
can be specified by --pipeline_config_path.

Example usage:
    ./train \
        --logtostderr \
        --train_dir=path/to/train_dir \
        --pipeline_config_path=pipeline_config.pbtxt

2) Three configuration files can be provided: a model_pb2.DetectionModel
configuration file to define what type of DetectionModel is being trained, an
input_reader_pb2.InputReader file to specify what training data will be used and
a train_pb2.TrainConfig file to configure training parameters.

Example usage:
    ./train \
        --logtostderr \
        --train_dir=path/to/train_dir \
        --model_config_path=model_config.pbtxt \
        --train_config_path=train_config.pbtxt \
        --input_config_path=train_input_config.pbtxt
"""

import functools
import json
import os
import tensorflow as tf

from object_detection import trainer
from object_detection.builders import dataset_builder
from object_detection.builders import graph_rewriter_builder
from object_detection.builders import model_builder
from object_detection.utils import config_util
from object_detection.utils import dataset_util

tf.logging.set_verbosity(tf.logging.INFO)

flags = tf.app.flags
flags.DEFINE_string('master', '', 'Name of the TensorFlow master to use.')
flags.DEFINE_integer('task', 0, 'task id')
flags.DEFINE_integer('num_clones', 1, 'Number of clones to deploy per worker.')
flags.DEFINE_boolean('clone_on_cpu', False,
                     'Force clones to be deployed on CPU.  Note that even if '
                     'set to False (allowing ops to run on gpu), some ops may '
                     'still be run on the CPU if they have no GPU kernel.')
flags.DEFINE_integer('worker_replicas', 1, 'Number of worker+trainer '
                     'replicas.')
flags.DEFINE_integer('ps_tasks', 0,
                     'Number of parameter server tasks. If None, does not use '
                     'a parameter server.')
flags.DEFINE_string('train_dir', '',
                    'Directory to save the checkpoints and training summaries.')

flags.DEFINE_string('pipeline_config_path', '',
                    'Path to a pipeline_pb2.TrainEvalPipelineConfig config '
                    'file. If provided, other configs are ignored')

flags.DEFINE_string('train_config_path', '',
                    'Path to a train_pb2.TrainConfig config file.')
flags.DEFINE_string('input_config_path', '',
                    'Path to an input_reader_pb2.InputReader config file.')
flags.DEFINE_string('model_config_path', '',
                    'Path to a model_pb2.DetectionModel config file.')

FLAGS = flags.FLAGS

def main(_):
  assert FLAGS.train_dir, '`train_dir` is missing.'
  if FLAGS.task == 0: tf.gfile.MakeDirs(FLAGS.train_dir)
  if FLAGS.pipeline_config_path:
    configs = config_util.get_configs_from_pipeline_file(
        FLAGS.pipeline_config_path)
    if FLAGS.task == 0:
      tf.gfile.Copy(FLAGS.pipeline_config_path,
                    os.path.join(FLAGS.train_dir, 'pipeline.config'),
                    overwrite=True)
  else:
    configs = config_util.get_configs_from_multiple_files(
        model_config_path=FLAGS.model_config_path,
        train_config_path=FLAGS.train_config_path,
        train_input_config_path=FLAGS.input_config_path)
    if FLAGS.task == 0:
      for name, config in [('model.config', FLAGS.model_config_path),
                           ('train.config', FLAGS.train_config_path),
                           ('input.config', FLAGS.input_config_path)]:
        tf.gfile.Copy(config, os.path.join(FLAGS.train_dir, name),
                      overwrite=True)

  model_config = configs['model']
  train_config = configs['train_config']
  input_config = configs['train_input_config']

  model_fn = functools.partial(
      model_builder.build,
      model_config=model_config,
      is_training=True)

  def get_next(config):
    return dataset_util.make_initializable_iterator(
        dataset_builder.build(config)).get_next()

  create_input_dict_fn = functools.partial(get_next, input_config)

  env = json.loads(os.environ.get('TF_CONFIG', '{}'))
  cluster_data = env.get('cluster', None)
  cluster = tf.train.ClusterSpec(cluster_data) if cluster_data else None
  task_data = env.get('task', None) or {'type': 'master', 'index': 0}
  task_info = type('TaskSpec', (object,), task_data)

  # Parameters for a single worker.
  ps_tasks = 0
  worker_replicas = 1
  worker_job_name = 'lonely_worker'
  task = 0
  is_chief = True
  master = ''

  if cluster_data and 'worker' in cluster_data:
    # Number of total worker replicas include "worker"s and the "master".
    worker_replicas = len(cluster_data['worker']) + 1
  if cluster_data and 'ps' in cluster_data:
    ps_tasks = len(cluster_data['ps'])

  if worker_replicas > 1 and ps_tasks < 1:
    raise ValueError('At least 1 ps task is needed for distributed training.')

  if worker_replicas >= 1 and ps_tasks > 0:
    # Set up distributed training.
    server = tf.train.Server(tf.train.ClusterSpec(cluster), protocol='grpc',
                             job_name=task_info.type,
                             task_index=task_info.index)
    if task_info.type == 'ps':
      server.join()
      return

    worker_job_name = '%s/task:%d' % (task_info.type, task_info.index)
    task = task_info.index
    is_chief = (task_info.type == 'master')
    master = server.target

  graph_rewriter_fn = None
  if 'graph_rewriter_config' in configs:
    graph_rewriter_fn = graph_rewriter_builder.build(
        configs['graph_rewriter_config'], is_training=True)

  trainer.train(
      create_input_dict_fn,
      model_fn,
      train_config,
      master,
      task,
      FLAGS.num_clones,
      worker_replicas,
      FLAGS.clone_on_cpu,
      ps_tasks,
      worker_job_name,
      is_chief,
      FLAGS.train_dir,
      graph_hook_fn=graph_rewriter_fn)

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

1、先定义了tf.app.flags，用于支持接受命令行传递参数，相当于接受argv。

flags = tf.app.flags
flags.DEFINE_string('master', '', 'Name of the TensorFlow master to use.')
flags.DEFINE_integer('task', 0, 'task id')
flags.DEFINE_integer('num_clones', 1, 'Number of clones to deploy per worker.')
flags.DEFINE_boolean('clone_on_cpu', False,
                     'Force clones to be deployed on CPU.  Note that even if '
                     'set to False (allowing ops to run on gpu), some ops may '
                     'still be run on the CPU if they have no GPU kernel.')
flags.DEFINE_integer('worker_replicas', 1, 'Number of worker+trainer '
                     'replicas.')
flags.DEFINE_integer('ps_tasks', 0,
                     'Number of parameter server tasks. If None, does not use '
                     'a parameter server.')
flags.DEFINE_string('train_dir', '',
                    'Directory to save the checkpoints and training summaries.')

flags.DEFINE_string('pipeline_config_path', '',
                    'Path to a pipeline_pb2.TrainEvalPipelineConfig config '
                    'file. If provided, other configs are ignored')

flags.DEFINE_string('train_config_path', '',
                    'Path to a train_pb2.TrainConfig config file.')
flags.DEFINE_string('input_config_path', '',
                    'Path to an input_reader_pb2.InputReader config file.')
flags.DEFINE_string('model_config_path', '',
                    'Path to a model_pb2.DetectionModel config file.')

FLAGS = flags.FLAGS

这里面有几个比较重要的参数，train_dir目录用于保存训练的模型和日志文件，pipeline_config_path用于指定pipeline_pb2.TrainEvalPipelineConfig配置文件的全路径(如果不指定指定这个参数，需要指定train_config_path，input_config_path，model_config_path配置文件，其实这三个文件就是把pipeline_pb2.TrainEvalPipelineConfig配置文件分成了三部分)。

2、再来看一下main函数，我们把它分成几部分来解读。

假设我们在控制台下的命令如下：

python train.py --train_dir voc/train_dir/ --pipeline_config_path voc/faster_rcnn_inception_resnet_v2_atrous_voc.config

3、第一部分

  assert FLAGS.train_dir, '`train_dir` is missing.'
  if FLAGS.task == 0: tf.gfile.MakeDirs(FLAGS.train_dir)
  if FLAGS.pipeline_config_path:
    configs = config_util.get_configs_from_pipeline_file(
        FLAGS.pipeline_config_path)
    if FLAGS.task == 0:
      tf.gfile.Copy(FLAGS.pipeline_config_path,
                    os.path.join(FLAGS.train_dir, 'pipeline.config'),
                    overwrite=True)
  else:
    configs = config_util.get_configs_from_multiple_files(
        model_config_path=FLAGS.model_config_path,
        train_config_path=FLAGS.train_config_path,
        train_input_config_path=FLAGS.input_config_path)
    if FLAGS.task == 0:
      for name, config in [('model.config', FLAGS.model_config_path),
                           ('train.config', FLAGS.train_config_path),
                           ('input.config', FLAGS.input_config_path)]:
        tf.gfile.Copy(config, os.path.join(FLAGS.train_dir, name),
                      overwrite=True)

因为我们传入了train_dir,pipeline_config_path参数，程序执行时会：

• 读取pipeline_config_path配置文件，返回一个dict，保存配置文件中`model`, `train_config`,  `train_input_config`, `eval_config`, `eval_input_config`信息。
• 把pipeline_config_path配置文件复制到train_dir目录下，命名为pipeline.config

4、第二部分

  model_config = configs['model']
  train_config = configs['train_config']
  input_config = configs['train_input_config']

  model_fn = functools.partial(
      model_builder.build,
      model_config=model_config,
      is_training=True)

  def get_next(config):
    return dataset_util.make_initializable_iterator(
        dataset_builder.build(config)).get_next()

  create_input_dict_fn = functools.partial(get_next, input_config)

• 变量model_config，train_config，input_config初始化
• model_builder.build函数，指定两个固定参数model_config，is_training并返回一个新的函数model_fn 。这个函数很重要，包括对目标检测模型的实现，后面会详细介绍。
• get_next函数，指定固定参数input_config。这个函数主要实现了tfrecord数据的读取，我们也放在后面介绍。

5、第三部分

  env = json.loads(os.environ.get('TF_CONFIG', '{}'))
  cluster_data = env.get('cluster', None)
  cluster = tf.train.ClusterSpec(cluster_data) if cluster_data else None
  task_data = env.get('task', None) or {'type': 'master', 'index': 0}
  task_info = type('TaskSpec', (object,), task_data)

  # Parameters for a single worker.
  ps_tasks = 0
  worker_replicas = 1
  worker_job_name = 'lonely_worker'
  task = 0
  is_chief = True
  master = ''

  if cluster_data and 'worker' in cluster_data:
    # Number of total worker replicas include "worker"s and the "master".
    worker_replicas = len(cluster_data['worker']) + 1
  if cluster_data and 'ps' in cluster_data:
    ps_tasks = len(cluster_data['ps'])

  if worker_replicas > 1 and ps_tasks < 1:
    raise ValueError('At least 1 ps task is needed for distributed training.')

  if worker_replicas >= 1 and ps_tasks > 0:
    # Set up distributed training.
    server = tf.train.Server(tf.train.ClusterSpec(cluster), protocol='grpc',
                             job_name=task_info.type,
                             task_index=task_info.index)
    if task_info.type == 'ps':
      server.join()
      return

    worker_job_name = '%s/task:%d' % (task_info.type, task_info.index)
    task = task_info.index
    is_chief = (task_info.type == 'master')
    master = server.target

• 这部分代码主要是用来实现分布式部署训练的。如果想了解的请点击这里第八节，配置分布式TensorFlow。

6、第四部分

  graph_rewriter_fn = None
  if 'graph_rewriter_config' in configs:
    graph_rewriter_fn = graph_rewriter_builder.build(
        configs['graph_rewriter_config'], is_training=True)

  trainer.train(
      create_input_dict_fn,
      model_fn,
      train_config,
      master,
      task,
      FLAGS.num_clones,
      worker_replicas,
      FLAGS.clone_on_cpu,
      ps_tasks,
      worker_job_name,
      is_chief,
      FLAGS.train_dir,
      graph_hook_fn=graph_rewriter_fn)

• 由于没有定义graph_rewriter_config，所以会直接执行trainer.train，开始读取数据，进行预处理后训练。

二  dataset_builder.build函数

先附上代码：

def build(input_reader_config, transform_input_data_fn=None,
          batch_size=None, max_num_boxes=None, num_classes=None,
          spatial_image_shape=None):
  """Builds a tf.data.Dataset.

  Builds a tf.data.Dataset by applying the `transform_input_data_fn` on all
  records. Applies a padded batch to the resulting dataset.

  Args:
    input_reader_config: A input_reader_pb2.InputReader object.
    transform_input_data_fn: Function to apply to all records, or None if
      no extra decoding is required.
    batch_size: Batch size. If None, batching is not performed.
    max_num_boxes: Max number of groundtruth boxes needed to compute shapes for
      padding. If None, will use a dynamic shape.
    num_classes: Number of classes in the dataset needed to compute shapes for
      padding. If None, will use a dynamic shape.
    spatial_image_shape: A list of two integers of the form [height, width]
      containing expected spatial shape of the image after applying
      transform_input_data_fn. If None, will use dynamic shapes.

  Returns:
    A tf.data.Dataset based on the input_reader_config.

  Raises:
    ValueError: On invalid input reader proto.
    ValueError: If no input paths are specified.
  """
  if not isinstance(input_reader_config, input_reader_pb2.InputReader):
    raise ValueError('input_reader_config not of type '
                     'input_reader_pb2.InputReader.')

  if input_reader_config.WhichOneof('input_reader') == 'tf_record_input_reader':
    config = input_reader_config.tf_record_input_reader
    if not config.input_path:
      raise ValueError('At least one input path must be specified in '
                       '`input_reader_config`.')

    label_map_proto_file = None
    if input_reader_config.HasField('label_map_path'):
      label_map_proto_file = input_reader_config.label_map_path

　　 #初始化需要解码的字段，以及解码对应字段的 handler
    decoder = tf_example_decoder.TfExampleDecoder(
        load_instance_masks=input_reader_config.load_instance_masks,
        instance_mask_type=input_reader_config.mask_type,
        label_map_proto_file=label_map_proto_file)

    def process_fn(value):
      processed = decoder.decode(value)
      if transform_input_data_fn is not None:
        return transform_input_data_fn(processed)
      return processed

　　 # 调用 tf.data.TFRecordDataset 从 config.input_path 读数据，调用 process_fn 对读取的数据解码数，预提取 input_reader_config.prefetch_size 条数据
    dataset = dataset_util.read_dataset(
        functools.partial(tf.data.TFRecordDataset, buffer_size=8 * 1000 * 1000),
        process_fn, config.input_path[:], input_reader_config)

    if batch_size:
      padding_shapes = _get_padding_shapes(dataset, max_num_boxes, num_classes,
                                           spatial_image_shape)
      dataset = dataset.apply(
          tf.contrib.data.padded_batch_and_drop_remainder(batch_size,
                                                          padding_shapes))
    return dataset

  raise ValueError('Unsupported input_reader_config.')

整个流程

• 获取训练集tfrecord文件路径，label_map_path文件路径，input_reader_config设置参数如下：

train_input_reader: {
  tf_record_input_reader {
    input_path: "voc/pascal_train.record"
  }
  label_map_path: "voc/pascal_label_map.pbtxt"
}

• 初始化需要解码的字段，以及解码对应字段的 handler
• 调用 tf.data.TFRecordDataset 从 config.input_path 读数据，调用 process_fn (定义了数据的解码格式)对读取的数据解码，预提取 input_reader_config.prefetch_size 条数据
• 对数据集应用 tf.contrib.data.padded_batch_and_drop_remainder，如果不够一个 batch_size 就丢弃该部分数据
• 返回一个迭代器

三 model_builder.build函数

代码如下：

def build(model_config, is_training, add_summaries=True,
          add_background_class=True):
  """Builds a DetectionModel based on the model config.

  Args:
    model_config: A model.proto object containing the config for the desired
      DetectionModel.
    is_training: True if this model is being built for training purposes.
    add_summaries: Whether to add tensorflow summaries in the model graph.
    add_background_class: Whether to add an implicit background class to one-hot
      encodings of groundtruth labels. Set to false if using groundtruth labels
      with an explicit background class or using multiclass scores instead of
      truth in the case of distillation. Ignored in the case of faster_rcnn.
  Returns:
    DetectionModel based on the config.

  Raises:
    ValueError: On invalid meta architecture or model.
  """
  if not isinstance(model_config, model_pb2.DetectionModel):
    raise ValueError('model_config not of type model_pb2.DetectionModel.')
  meta_architecture = model_config.WhichOneof('model')
  if meta_architecture == 'ssd':
    return _build_ssd_model(model_config.ssd, is_training, add_summaries,
                            add_background_class)
  if meta_architecture == 'faster_rcnn':
    return _build_faster_rcnn_model(model_config.faster_rcnn, is_training,
                                    add_summaries)
  raise ValueError('Unknown meta architecture: {}'.format(meta_architecture))

先获取我们使用的目标检测模型，由于我们使用的是faster_rcnn_inception_resnet_v2，因此会调用_build_faster_rcnn_model函数，并且传入参数faster_rcnn，is_training,add_summaries。其中faster_rcnn的内容如下：

model {
  faster_rcnn {
    num_classes:
    image_resizer {
      keep_aspect_ratio_resizer {
        min_dimension:
        max_dimension:
      }
    }
    feature_extractor {
      type: 'faster_rcnn_inception_resnet_v2'
      first_stage_features_stride:
    }
    first_stage_anchor_generator {
      grid_anchor_generator {
        scales: [0.25, 0.5, 1.0, 2.0]
        aspect_ratios: [0.5, 1.0, 2.0]
        height_stride:
        width_stride:
      }
    }
    first_stage_atrous_rate:
    first_stage_box_predictor_conv_hyperparams {
      op: CONV
      regularizer {
        l2_regularizer {
          weight: 0.0
        }
      }
      initializer {
        truncated_normal_initializer {
          stddev: 0.01
        }
      }
    }
    first_stage_nms_score_threshold: 0.0
    first_stage_nms_iou_threshold: 0.7
    first_stage_max_proposals:
    first_stage_localization_loss_weight: 2.0
    first_stage_objectness_loss_weight: 1.0
    initial_crop_size:
    maxpool_kernel_size:
    maxpool_stride:
    second_stage_box_predictor {
      mask_rcnn_box_predictor {
        use_dropout: false
        dropout_keep_probability: 1.0
        fc_hyperparams {
          op: FC
          regularizer {
            l2_regularizer {
              weight: 0.0
            }
          }
          initializer {
            variance_scaling_initializer {
              factor: 1.0
              uniform: true
              mode: FAN_AVG
            }
          }
        }
      }
    }
    second_stage_post_processing {
      batch_non_max_suppression {
        score_threshold: 0.0
        iou_threshold: 0.6
        max_detections_per_class:
        max_total_detections:
      }
      score_converter: SOFTMAX
    }
    second_stage_localization_loss_weight: 2.0
    second_stage_classification_loss_weight: 1.0
  }
}

我们再来看一下_build_faster_rcnn_model的源码：

def _build_faster_rcnn_model(frcnn_config, is_training, add_summaries):
  """Builds a Faster R-CNN or R-FCN detection model based on the model config.

  Builds R-FCN model if the second_stage_box_predictor in the config is of type
  `rfcn_box_predictor` else builds a Faster R-CNN model.

  Args:
    frcnn_config: A faster_rcnn.proto object containing the config for the
      desired FasterRCNNMetaArch or RFCNMetaArch.
    is_training: True if this model is being built for training purposes.
    add_summaries: Whether to add tf summaries in the model.

  Returns:
    FasterRCNNMetaArch based on the config.

  Raises:
    ValueError: If frcnn_config.type is not recognized (i.e. not registered in
      model_class_map).
  """
  num_classes = frcnn_config.num_classes
  image_resizer_fn = image_resizer_builder.build(frcnn_config.image_resizer)

  feature_extractor = _build_faster_rcnn_feature_extractor(
      frcnn_config.feature_extractor, is_training,
      frcnn_config.inplace_batchnorm_update)

  number_of_stages = frcnn_config.number_of_stages
  first_stage_anchor_generator = anchor_generator_builder.build(
      frcnn_config.first_stage_anchor_generator)

  first_stage_atrous_rate = frcnn_config.first_stage_atrous_rate
  first_stage_box_predictor_arg_scope_fn = hyperparams_builder.build(
      frcnn_config.first_stage_box_predictor_conv_hyperparams, is_training)
  first_stage_box_predictor_kernel_size = (
      frcnn_config.first_stage_box_predictor_kernel_size)
  first_stage_box_predictor_depth = frcnn_config.first_stage_box_predictor_depth
  first_stage_minibatch_size = frcnn_config.first_stage_minibatch_size
  first_stage_positive_balance_fraction = (
      frcnn_config.first_stage_positive_balance_fraction)
  first_stage_nms_score_threshold = frcnn_config.first_stage_nms_score_threshold
  first_stage_nms_iou_threshold = frcnn_config.first_stage_nms_iou_threshold
  first_stage_max_proposals = frcnn_config.first_stage_max_proposals
  first_stage_loc_loss_weight = (
      frcnn_config.first_stage_localization_loss_weight)
  first_stage_obj_loss_weight = frcnn_config.first_stage_objectness_loss_weight

  initial_crop_size = frcnn_config.initial_crop_size
  maxpool_kernel_size = frcnn_config.maxpool_kernel_size
  maxpool_stride = frcnn_config.maxpool_stride

  second_stage_box_predictor = box_predictor_builder.build(
      hyperparams_builder.build,
      frcnn_config.second_stage_box_predictor,
      is_training=is_training,
      num_classes=num_classes)
  second_stage_batch_size = frcnn_config.second_stage_batch_size
  second_stage_balance_fraction = frcnn_config.second_stage_balance_fraction
  (second_stage_non_max_suppression_fn, second_stage_score_conversion_fn
  ) = post_processing_builder.build(frcnn_config.second_stage_post_processing)
  second_stage_localization_loss_weight = (
      frcnn_config.second_stage_localization_loss_weight)
  second_stage_classification_loss = (
      losses_builder.build_faster_rcnn_classification_loss(
          frcnn_config.second_stage_classification_loss))
  second_stage_classification_loss_weight = (
      frcnn_config.second_stage_classification_loss_weight)
  second_stage_mask_prediction_loss_weight = (
      frcnn_config.second_stage_mask_prediction_loss_weight)

  hard_example_miner = None
  if frcnn_config.HasField('hard_example_miner'):
    hard_example_miner = losses_builder.build_hard_example_miner(
        frcnn_config.hard_example_miner,
        second_stage_classification_loss_weight,
        second_stage_localization_loss_weight)

  common_kwargs = {
      'is_training': is_training,
      'num_classes': num_classes,
      'image_resizer_fn': image_resizer_fn,
      'feature_extractor': feature_extractor,
      'number_of_stages': number_of_stages,
      'first_stage_anchor_generator': first_stage_anchor_generator,
      'first_stage_atrous_rate': first_stage_atrous_rate,
      'first_stage_box_predictor_arg_scope_fn':
      first_stage_box_predictor_arg_scope_fn,
      'first_stage_box_predictor_kernel_size':
      first_stage_box_predictor_kernel_size,
      'first_stage_box_predictor_depth': first_stage_box_predictor_depth,
      'first_stage_minibatch_size': first_stage_minibatch_size,
      'first_stage_positive_balance_fraction':
      first_stage_positive_balance_fraction,
      'first_stage_nms_score_threshold': first_stage_nms_score_threshold,
      'first_stage_nms_iou_threshold': first_stage_nms_iou_threshold,
      'first_stage_max_proposals': first_stage_max_proposals,
      'first_stage_localization_loss_weight': first_stage_loc_loss_weight,
      'first_stage_objectness_loss_weight': first_stage_obj_loss_weight,
      'second_stage_batch_size': second_stage_batch_size,
      'second_stage_balance_fraction': second_stage_balance_fraction,
      'second_stage_non_max_suppression_fn':
      second_stage_non_max_suppression_fn,
      'second_stage_score_conversion_fn': second_stage_score_conversion_fn,
      'second_stage_localization_loss_weight':
      second_stage_localization_loss_weight,
      'second_stage_classification_loss':
      second_stage_classification_loss,
      'second_stage_classification_loss_weight':
      second_stage_classification_loss_weight,
      'hard_example_miner': hard_example_miner,
      'add_summaries': add_summaries}

  if isinstance(second_stage_box_predictor, box_predictor.RfcnBoxPredictor):
    return rfcn_meta_arch.RFCNMetaArch(
        second_stage_rfcn_box_predictor=second_stage_box_predictor,
        **common_kwargs)
  else:
    return faster_rcnn_meta_arch.FasterRCNNMetaArch(
        initial_crop_size=initial_crop_size,
        maxpool_kernel_size=maxpool_kernel_size,
        maxpool_stride=maxpool_stride,
        second_stage_mask_rcnn_box_predictor=second_stage_box_predictor,
        second_stage_mask_prediction_loss_weight=(
            second_stage_mask_prediction_loss_weight),
        **common_kwargs)

• 代码的前半部分是加载Faster  R-CNN模型的参数，如num_classes，第一阶段RPN网络的参数，第二阶段Fast R-CNN网络的参数。
• 由于配置文件中指定了网络结构是faster_rcnn_inception_resnet_v2，因此会调用inception_resnet_v2网络进行特征提取(位于object_detection\object_detection\models文件夹下)，特征提取分为两部分，利用inception_resnet_v2的前半部分得到供RPN网络输入的特征图，网络继续前向传播至特有卷积层，产生更高维特征图，用于Fast R-CNN网络网络的特征提取，详情参考第三十一节，目标检测算法之 Faster R-CNN算法详解。
• 代码后半部分用来构建目标检测模型以及损失函数。

四 trainer.train函数

代码如下：

def train(create_tensor_dict_fn,
          create_model_fn,
          train_config,
          master,
          task,
          num_clones,
          worker_replicas,
          clone_on_cpu,
          ps_tasks,
          worker_job_name,
          is_chief,
          train_dir,
          graph_hook_fn=None):
  """Training function for detection models.

  Args:
    create_tensor_dict_fn: a function to create a tensor input dictionary.
    create_model_fn: a function that creates a DetectionModel and generates
                     losses.
    train_config: a train_pb2.TrainConfig protobuf.
    master: BNS name of the TensorFlow master to use.
    task: The task id of this training instance.
    num_clones: The number of clones to run per machine.
    worker_replicas: The number of work replicas to train with.
    clone_on_cpu: True if clones should be forced to run on CPU.
    ps_tasks: Number of parameter server tasks.
    worker_job_name: Name of the worker job.
    is_chief: Whether this replica is the chief replica.
    train_dir: Directory to write checkpoints and training summaries to.
    graph_hook_fn: Optional function that is called after the inference graph is
      built (before optimization). This is helpful to perform additional changes
      to the training graph such as adding FakeQuant ops. The function should
      modify the default graph.
  """

  detection_model = create_model_fn()
  data_augmentation_options = [
      preprocessor_builder.build(step)
      for step in train_config.data_augmentation_options]

  with tf.Graph().as_default():
    # Build a configuration specifying multi-GPU and multi-replicas.
    deploy_config = model_deploy.DeploymentConfig(
        num_clones=num_clones,
        clone_on_cpu=clone_on_cpu,
        replica_id=task,
        num_replicas=worker_replicas,
        num_ps_tasks=ps_tasks,
        worker_job_name=worker_job_name)

    # Place the global step on the device storing the variables.
    with tf.device(deploy_config.variables_device()):
      global_step = slim.create_global_step()

    with tf.device(deploy_config.inputs_device()):
      input_queue = create_input_queue(
          train_config.batch_size // num_clones, create_tensor_dict_fn,
          train_config.batch_queue_capacity,
          train_config.num_batch_queue_threads,
          train_config.prefetch_queue_capacity, data_augmentation_options)

    # Gather initial summaries.
    # TODO(rathodv): See if summaries can be added/extracted from global tf
    # collections so that they don't have to be passed around.
    summaries = set(tf.get_collection(tf.GraphKeys.SUMMARIES))
    global_summaries = set([])

    model_fn = functools.partial(_create_losses,
                                 create_model_fn=create_model_fn,
                                 train_config=train_config)
    clones = model_deploy.create_clones(deploy_config, model_fn, [input_queue])
    first_clone_scope = clones[0].scope

    if graph_hook_fn:
      with tf.device(deploy_config.variables_device()):
        graph_hook_fn()

    # Gather update_ops from the first clone. These contain, for example,
    # the updates for the batch_norm variables created by model_fn.
    update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS, first_clone_scope)

    with tf.device(deploy_config.optimizer_device()):
      training_optimizer, optimizer_summary_vars = optimizer_builder.build(
          train_config.optimizer)
      for var in optimizer_summary_vars:
        tf.summary.scalar(var.op.name, var, family='LearningRate')

    sync_optimizer = None
    if train_config.sync_replicas:
      training_optimizer = tf.train.SyncReplicasOptimizer(
          training_optimizer,
          replicas_to_aggregate=train_config.replicas_to_aggregate,
          total_num_replicas=worker_replicas)
      sync_optimizer = training_optimizer

    with tf.device(deploy_config.optimizer_device()):
      regularization_losses = (None if train_config.add_regularization_loss
                               else [])
      total_loss, grads_and_vars = model_deploy.optimize_clones(
          clones, training_optimizer,
          regularization_losses=regularization_losses)
      total_loss = tf.check_numerics(total_loss, 'LossTensor is inf or nan.')

      # Optionally multiply bias gradients by train_config.bias_grad_multiplier.
      if train_config.bias_grad_multiplier:
        biases_regex_list = ['.*/biases']
        grads_and_vars = variables_helper.multiply_gradients_matching_regex(
            grads_and_vars,
            biases_regex_list,
            multiplier=train_config.bias_grad_multiplier)

      # Optionally freeze some layers by setting their gradients to be zero.
      if train_config.freeze_variables:
        grads_and_vars = variables_helper.freeze_gradients_matching_regex(
            grads_and_vars, train_config.freeze_variables)

      # Optionally clip gradients
      if train_config.gradient_clipping_by_norm > 0:
        with tf.name_scope('clip_grads'):
          grads_and_vars = slim.learning.clip_gradient_norms(
              grads_and_vars, train_config.gradient_clipping_by_norm)

      # Create gradient updates.
      grad_updates = training_optimizer.apply_gradients(grads_and_vars,
                                                        global_step=global_step)
      update_ops.append(grad_updates)
      update_op = tf.group(*update_ops, name='update_barrier')
      with tf.control_dependencies([update_op]):
        train_tensor = tf.identity(total_loss, name='train_op')

    # Add summaries.
    for model_var in slim.get_model_variables():
      global_summaries.add(tf.summary.histogram('ModelVars/' +
                                                model_var.op.name, model_var))
    for loss_tensor in tf.losses.get_losses():
      global_summaries.add(tf.summary.scalar('Losses/' + loss_tensor.op.name,
                                             loss_tensor))
    global_summaries.add(
        tf.summary.scalar('Losses/TotalLoss', tf.losses.get_total_loss()))

    # Add the summaries from the first clone. These contain the summaries
    # created by model_fn and either optimize_clones() or _gather_clone_loss().
    summaries |= set(tf.get_collection(tf.GraphKeys.SUMMARIES,
                                       first_clone_scope))
    summaries |= global_summaries

    # Merge all summaries together.
    summary_op = tf.summary.merge(list(summaries), name='summary_op')

    # Soft placement allows placing on CPU ops without GPU implementation.
    session_config = tf.ConfigProto(allow_soft_placement=True,
                                    log_device_placement=False)

    # Save checkpoints regularly.
    keep_checkpoint_every_n_hours = train_config.keep_checkpoint_every_n_hours
    saver = tf.train.Saver(
        keep_checkpoint_every_n_hours=keep_checkpoint_every_n_hours)

    # Create ops required to initialize the model from a given checkpoint.
    init_fn = None
    if train_config.fine_tune_checkpoint:
      if not train_config.fine_tune_checkpoint_type:
        # train_config.from_detection_checkpoint field is deprecated. For
        # backward compatibility, fine_tune_checkpoint_type is set based on
        # from_detection_checkpoint.
        if train_config.from_detection_checkpoint:
          train_config.fine_tune_checkpoint_type = 'detection'
        else:
          train_config.fine_tune_checkpoint_type = 'classification'
      var_map = detection_model.restore_map(
          fine_tune_checkpoint_type=train_config.fine_tune_checkpoint_type,
          load_all_detection_checkpoint_vars=(
              train_config.load_all_detection_checkpoint_vars))
      available_var_map = (variables_helper.
                           get_variables_available_in_checkpoint(
                               var_map, train_config.fine_tune_checkpoint))
      init_saver = tf.train.Saver(available_var_map)
      def initializer_fn(sess):
        init_saver.restore(sess, train_config.fine_tune_checkpoint)
      init_fn = initializer_fn

    slim.learning.train(
        train_tensor,
        logdir=train_dir,
        master=master,
        is_chief=is_chief,
        session_config=session_config,
        startup_delay_steps=train_config.startup_delay_steps,
        init_fn=init_fn,
        summary_op=summary_op,
        number_of_steps=(
            train_config.num_steps if train_config.num_steps else None),
        save_summaries_secs=120,
        sync_optimizer=sync_optimizer,
        saver=saver)

• 批量读数据前，通过 data_augmentation_options (数据增强)类指定预处理操作

• 批量数据读取：创建两个队列
  队列1 ： 开启 N 个线程，每个线程从数据集依次读一条数据，写入队列 1。一个线程从队列 1 每次读 batch_size 条数据
  队列2：将队列 1 出队列的数据写入队列 2， 当调用 dequeue 的事实，从队列 2 读取 batch_size 的数据。
• 定义损失函数，优化器，Saver对象。
• 批量读数据后，通过模型的预处理函数进行预处理 detection_model.preprocess 之后，喂给模型，开始训练。

五 总结

一个训练流程下来，我们会发现谷歌源代码封装的很严重，不易于我们之后的改进，因此自己实现一个目标检测模型显得尤为重要，后面我会简单的实现一个目标检测算法。

 参考文章

[1]Tensorflow 物体识别(object detection) 之如何优雅地读数据

[2]Tensorflow学习笔记-通过slim读取TFRecord文件

[3]Tensorflow 物体识别(object detection) 之如何优雅地预处理数据