:mod:`blueoil.networks.segmentation.lm_bisenet`
===============================================

.. py:module:: blueoil.networks.segmentation.lm_bisenet


Module Contents
---------------

Classes
~~~~~~~

.. autoapisummary::

   blueoil.networks.segmentation.lm_bisenet.LMBiSeNet
   blueoil.networks.segmentation.lm_bisenet.LMBiSeNetQuantize



.. py:class:: LMBiSeNet(weight_decay_rate=0.0, auxiliary_loss_weight=0.5, use_feature_fusion=True, use_attention_refinement=True, use_tail_gap=True, *args, **kwargs)

   Bases: :class:`blueoil.networks.segmentation.base.Base`

   LM original semantic segmentation network reference to [BiSeNet](https://arxiv.org/abs/1808.00897)

   Major difference from BiSeNet:
   * Apply first convolution then branch into contextual and spatial path.
   * Use space to depth (s2d) and depth to space (d2s) as upsample and downsample.
   * Use only 1 stride 1x1 and 3x3 convolution (not use multi stride and dilated convolution)
       for be limited by our convolution IP.
   * Use DensNet block in contextual part.
   * All of convolution out channels are less than BiSeNet for inference time.
   * Use attention refinement module reference to BiSeNet after last layer of 1/32
       (BiSeNet: both 1/16 and 1/32) in context path.
   * Use relu activation instead of sigmoid in attention refinement and feature fusion module.
   * In up-sampling followed by context path, alternate d2s and 1x1 conv for reducing channel size.

   .. method:: _space_to_depth(self, name, inputs=None, block_size=2)


   .. method:: _depth_to_space(self, name, inputs=None, block_size=2)


   .. method:: _batch_norm(self, inputs, training)


   .. method:: _conv_bias(self, name, inputs, filters, kernel_size)


   .. method:: _spatial(self, x)


   .. method:: _context(self, x)


   .. method:: _attention(self, name, x)


   .. method:: _fusion(self, sp, cx)

      Feature fusion module


   .. method:: base(self, images, is_training, *args, **kwargs)

      Base function contains inference.

      :param images: Input images.
      :param is_training: A flag for if is training.

      :returns: Inference result.
      :rtype: tf.Tensor


   .. method:: _cross_entropy(self, x, labels)


   .. method:: _weight_decay_loss(self)

      L2 weight decay (regularization) loss.


   .. method:: loss(self, output, labels)

      Loss.

      :param output: tensor from inference.
      :param labels: labels tensor.


   .. method:: summary(self, output, labels=None)

      Summary.

      :param output: tensor from inference.
      :param labels: labels tensor.


   .. method:: metrics(self, output, labels)

      Metrics.

      :param output: tensor from inference.
      :param labels: labels tensor.


   .. method:: post_process(self, output)



.. py:class:: LMBiSeNetQuantize(activation_quantizer=None, activation_quantizer_kwargs={}, weight_quantizer=None, weight_quantizer_kwargs={}, *args, **kwargs)

   Bases: :class:`blueoil.networks.segmentation.lm_bisenet.LMBiSeNet`

   Following `args` are used for inference: ``activation_quantizer``, ``activation_quantizer_kwargs``,
      ``weight_quantizer``, ``weight_quantizer_kwargs``.

   :param activation_quantizer: Weight quantizer. See more at `blueoil.quantizations`.
   :type activation_quantizer: callable
   :param activation_quantizer_kwargs: Kwargs for `activation_quantizer`.
   :type activation_quantizer_kwargs: dict
   :param weight_quantizer: Activation quantizer. See more at `blueoil.quantizations`.
   :type weight_quantizer: callable
   :param weight_quantizer_kwargs: Kwargs for `weight_quantizer`.
   :type weight_quantizer_kwargs: dict

   .. method:: _quantized_variable_getter(getter, name, weight_quantization=None, quantize_first_convolution=False, *args, **kwargs)
      :staticmethod:

      Get the quantized variables.

      Use if to choose or skip the target should be quantized.

      :param getter: Default from tensorflow.
      :param name: Default from tensorflow.
      :param weight_quantization: Callable object which quantize variable.
      :param args: Args.
      :param kwargs: Kwargs.


   .. method:: base(self, images, is_training)

      Base function contains inference.

      :param images: Input images.
      :param is_training: A flag for if is training.

      :returns: Inference result.
      :rtype: tf.Tensor