:mod:`blueoil.networks.base`
============================

.. py:module:: blueoil.networks.base


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

Classes
~~~~~~~

.. autoapisummary::

   blueoil.networks.base.BaseNetwork



.. py:class:: BaseNetwork(is_debug=False, optimizer_class=tf.compat.v1.train.GradientDescentOptimizer, optimizer_kwargs=None, learning_rate_func=None, learning_rate_kwargs=None, classes=(), image_size=(), batch_size=64, data_format='NHWC')

   Bases: :class:`object`

   Base network.

   This base network is for every task, such as classification, object detection and segmentation.
   Every sub task's base network class should extend this class.

   :param is_debug: Set True to use debug mode. It will summary some histograms,
                    use small dataset and step size.
   :type is_debug: boolean
   :param optimizer_class: Optimizer using for training.
   :type optimizer_class: class
   :param optimizer_kwargs: For init optimizer.
   :type optimizer_kwargs: dict
   :param learning_rate_func: Use for changing learning rate. Such as learning rate decay,
                              `tf.train.piecewise_constant`.
   :type learning_rate_func: callable
   :param learning_rate_kwargs: For learning rate function. For example of `tf.train.piecewise_constant`,
                                `{"values": [5e-5, 1e-5, 5e-6, 1e-6, 5e-7], "boundaries": [20000, 40000, 60000, 80000]}`.
   :type learning_rate_kwargs: dict
   :param classes: Classes names list.
   :type classes: list | tuple
   :param image_size: Image size.
   :type image_size: list | tuple
   :param batch_size: Batch size.
   :type batch_size: list | tuple

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

      Base function contains inference.

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

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


   .. method:: placeholders(self)
      :abstractmethod:

      Placeholders.

      Return placeholders.

      :returns: Placeholders.
      :rtype: tf.compat.v1.placeholder


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

      Metrics.

      :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:: inference(self, images, is_training)
      :abstractmethod:

      Inference.

      :param images: images tensor. shape is (batch_num, height, width, channel)


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

      Loss.

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


   .. method:: optimizer(self)


   .. method:: train(self, loss, optimizer, var_list=[])

      Train.

      :param loss: loss function of this network.