DeepLabv3+代码阅读之common.py

一、ModelOptions

class ModelOptions(
    collections.namedtuple('ModelOptions', [
        'outputs_to_num_classes',# 输出类型对应类别数的字典，会得到如：outputs_to_num_classes['semantic'] = 21
        'crop_size',# 裁剪尺寸
        'atrous_rates',# 对于`xception_65`，output_stride = 8，atrous_rates = [12, 24, 36]
        'output_stride',# 16
        'preprocessed_images_dtype',
        'merge_method',
        'add_image_level_feature',
        'image_pooling_crop_size',
        'image_pooling_stride',
        'aspp_with_batch_norm',
        'aspp_with_separable_conv',
        'multi_grid',
        'decoder_output_stride',
        'decoder_use_separable_conv',
        'logits_kernel_size',
        'model_variant',
        'depth_multiplier',
        'divisible_by',
        'prediction_with_upsampled_logits',
        'dense_prediction_cell_config',
        'nas_stem_output_num_conv_filters',
        'use_bounded_activation'
    ])):
  """Immutable class to hold model options."""

  __slots__ = ()

  def __new__(cls,
              outputs_to_num_classes,
              crop_size=None,
              atrous_rates=None,
              output_stride=8,
              preprocessed_images_dtype=tf.float32):
    """Constructor to set default values.

    Args:
      outputs_to_num_classes: A dictionary from output type to the number of
        classes. For example, for the task of semantic segmentation with 21
        semantic classes, we would have outputs_to_num_classes['semantic'] = 21.
      crop_size: A tuple [crop_height, crop_width].
      atrous_rates: A list of atrous convolution rates for ASPP.
      output_stride: The ratio of input to output spatial resolution.
      preprocessed_images_dtype: The type after the preprocessing function.

    Returns:
      A new ModelOptions instance.
    """
    dense_prediction_cell_config = None
    if FLAGS.dense_prediction_cell_json:
      with tf.gfile.Open(FLAGS.dense_prediction_cell_json, 'r') as f:
        dense_prediction_cell_config = json.load(f)
    decoder_output_stride = None
    if FLAGS.decoder_output_stride:
      decoder_output_stride = [
          int(x) for x in FLAGS.decoder_output_stride]
      if sorted(decoder_output_stride, reverse=True) != decoder_output_stride:
        raise ValueError('Decoder output stride need to be sorted in the '
                         'descending order.')
    image_pooling_crop_size = None
    if FLAGS.image_pooling_crop_size:
      image_pooling_crop_size = [int(x) for x in FLAGS.image_pooling_crop_size]
    image_pooling_stride = [1, 1]
    if FLAGS.image_pooling_stride:
      image_pooling_stride = [int(x) for x in FLAGS.image_pooling_stride]
    return super(ModelOptions, cls).__new__(
        cls, outputs_to_num_classes, crop_size, atrous_rates, output_stride,
        preprocessed_images_dtype, FLAGS.merge_method,
        FLAGS.add_image_level_feature,
        image_pooling_crop_size,
        image_pooling_stride,
        FLAGS.aspp_with_batch_norm,
        FLAGS.aspp_with_separable_conv, FLAGS.multi_grid, decoder_output_stride,
        FLAGS.decoder_use_separable_conv, FLAGS.logits_kernel_size,
        FLAGS.model_variant, FLAGS.depth_multiplier, FLAGS.divisible_by,
        FLAGS.prediction_with_upsampled_logits, dense_prediction_cell_config,
        FLAGS.nas_stem_output_num_conv_filters, FLAGS.use_bounded_activation)

  def __deepcopy__(self, memo):
    return ModelOptions(copy.deepcopy(self.outputs_to_num_classes),
                        self.crop_size,
                        self.atrous_rates,
                        self.output_stride,
                        self.preprocessed_images_dtype)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85