Modules Available By Default

With this library, objects are created based on the class names and parameters specified in the config files or at the command line. For example, shufflenet_v2_x1_0 is one of the models in the torchvision.models module, so you can use it for your experiment like this:

In a config file:

models:
  trunk:
    shufflenet_v2_x1_0:
      pretrained: True

At the command line:

--models~OVERRIDE~ {trunk: {shufflenet_v2_x1_0: {pretrained: True}}}

By default, the following modules are available.

• Models
  • pretrainedmodels
  • torchvision.models
• Optimizers
  • torch.optim
  • torch.optim.lr_scheduler
• Datasets
  • powerful_benchmarker.datasets
  • torchvision.datasets
• Transforms
  • torchvision.transforms.transforms
  • easy_module_attribute_getter.custom_transforms
• Losses
  • pytorch_metric_learning.losses
  • torch.nn
• Miners
  • pytorch_metric_learning.miners
• Samplers
  • pytorch_metric_learning.samplers
  • torch.utils.data
• Trainers
  • pytorch_metric_learning.trainers
• Testers
  • pytorch_metric_learning.testers

You can add other classes and modules by using the register functionality.