ImageRepresentation: Code Skeleton

addict.Dict

Class which implements a dictionary that provides attribute-style access.
This class is used to implement configurations of all morphosearch classes, typically initialized in the class __init__ function with:

self.config = self.__class__.default_config()
self.config.update(config)
self.config.update(kwargs)

image_representation.Representation

The main API methods that this class needs to implement are:

• calc_embedding(x): given an input image x calculate the embedding
• save(filepath): saves the model

image_representation.TorchNNRepresentation

Base class of representations that are also torch neural modules. Inherits from image_representation.Representation and torch.nn.Module.

• config:
  • config.network:
    • config.network.name:
    • config.network.parameters:
    • config.network.weights_init:
    • config.network.weights_init.name:
    • config.network.weights_init.parameters:
  • config.device: 'cpu', 'cuda'
  • config.loss:
    • config.loss.name
    • config.loss.parameters
  • config.optimizer:
    • config.optimizer.name
    • config.optimizer.parameters
  • config.logging:
  • config.checkpoint:
    • config.checkpoint.folder:
• network: torch.nn.Module or Dict of torch.nn.Modules with several sub networks (encoder, decoder, etc)
• loss_f: torch.nn.functional or Dict of sub losses(discriminator, generator)
• optimizer: torch.optim or Dict of sub optimizers (discriminator, generator)
• n_epochs: number of training epochs
• n_latents: number of dimensions of the encoding

Aditionnally to Representation's main API methods, the following main API methods must be implemented:

• set_network(network_name, network_parameters):
• init_network(weights_init_name, weights_init_parameters):
• set_loss(loss_name, loss_parameters):
• set_optimizer (optimizer_name, optimizer_parameters):
• run_training (train_loader, n_epochs, valid_loader = None, training_logger=None):
• train_epoch (train_loader, logger = None):
• valid_epoch (valid_loader, logger = None):
• save(filepath):
• load(filepath, map_loaction='cpu'):
• calc_embedding (x): given an input image x calc the embedding z