segmentation_application.py

Segmentation Application Class

What do I know?

• Segmentation Application inherits from BaseApplication

  • BaseApplication has instance variables

    Instance Variable	Default Value	Perceived Purpose
    REQUIRED_CONFIG_SECTION	None	defines name of customized configuration file section
    SUPPORTED_PHASES	{TRAIN, INFER, EVAL}	Suggests the types of things we can run on our data
    _action	TRAIN	The current action of the network (either train, inference, or evaluation)
    is_validation	None	Allows switching between validation and training (which are mutually exclusive)
    readers	None	Defines how to read data
    sampler	None	Uses readers as input to generate sequences of ImageWindow that will be fed into the networks
    net	None	training the network
    optimiser	None	training the network
    gradient_op	None	interpret network output
    output_decoder	None	interpret network output
    gradients_collector	None	interpret network output

    • BaseApplication requires the following functions to be implemented

      • initialize_dataset_loader
      • initialize_sampler
      • initialise_network
      • connect_data_and_network
      • interpret_output

    • SegmentationApplication also adds its own instance variables

    Instance Variable	Default Value	Perceived Purpose
    net_param	Set by parameter	Contains many parameters for the model including preprocessing layers parameters, and initializing samplers
    action_param	Set by parameter	Used in the initialization of aggregators and samplers. Contains parameters for how the data is processed such as sample per volume, validation every n, etc.
    data_param	None	Sets the window sizes of the sampler
    segmentation_param	None	Sets the number of classes the data is to be segmented into
    SUPPORTED_SAMPLING	'uniform', 'weighted', 'resize', 'balanced'	Initializes and lists the different way data can be sampled