Tensorflow开源的object detection API中的源码解析（三）：faster_rcnn_inception_resnet_v2_feature_extractor.py

对于faster_rcnn_meta_arch.py的理解见这篇文章
对于inception_resnet_v2.py的理解见这篇文章

"""Inception Resnet v2 Faster R-CNN implementation.

参见Szegedy等人的“Inception-v4，Inception-ResNet Impact of Residual Connections on
Learning"（论文地址：https://arxiv.org/abs/1602.07261）
以及Huang等人的
“Speed/accuracy trade-offs for modern convolutional object detectors" by
（论文地址：https://arxiv.org/abs/1611.10012）
"""

import tensorflow as tf

from object_detection.meta_architectures import faster_rcnn_meta_arch
from object_detection.net import inception_resnet_v2

slim = tf.contrib.slim


class FasterRCNNInceptionResnetV2FeatureExtractor(
    faster_rcnn_meta_arch.FasterRCNNFeatureExtractor):
  """ 这个类主要是Inception_Resnet_v2 Faster R-CNN的特征提取器的相关实现 """

  def __init__(self,
               is_training,
               first_stage_features_stride,
               reuse_weights=None,
               weight_decay=0.0):
    """Constructor.

    Args:
      is_training: 见faster_rcnn_meta_arch的构造器
      first_stage_features_stride: 同上
      reuse_weights: 同上
      weight_decay: 同上

    Raises:
      ValueError: 如果 `first_stage_features_stride` 既不是8也不是16。
    """
    if first_stage_features_stride != 8 and first_stage_features_stride != 16:
      raise ValueError('`first_stage_features_stride` must be 8 or 16.')
    super(FasterRCNNInceptionResnetV2FeatureExtractor, self).__init__(
        is_training, first_stage_features_stride, reuse_weights, weight_decay)

  def preprocess