TrainConfig

class plugins.train.trainer.base.TrainConfig(folders: list[str], batch_size: int, augment_color: bool, flip: bool, warp: bool, cache_landmarks: bool, lr_finder: bool = False, snapshot_interval: int = -1)

Bases: object

Configuration for training a model

Parameters:

  • image_folders – List of folders to be used as inputs to the model. Folders are provided in processing order (eg: [A, B, …])

  • batch_size (int) – The batch size to load data from each of the loaders

  • augment_color (bool) – True to perform color augmentation otherwise False

  • flip (bool) – True to perform image flipping otherwise False

  • warp (bool) – False to disable warping True to enable warping

  • cache_landmarks (bool) – True to cache landmarks from the other side for Warp to landmarks

  • use_lr_finder – True to use the learning rate finder. Default: False

  • interval (snapshot) – The number of iterations between snapshots. Default -1 (Disabled)

  • folders (list[str])

  • lr_finder (bool)

  • snapshot_interval (int)

Attributes Summary

lr_finder

True to use the learning rate finder

snapshot_interval

The number of iterations between snapshots

Attributes Documentation

lr_finder: bool = False

True to use the learning rate finder

snapshot_interval: int = -1

The number of iterations between snapshots

augment_color: bool = <dataclasses._MISSING_TYPE object>

True to perform color augmentation otherwise False

batch_size: int = <dataclasses._MISSING_TYPE object>

The batch size to load data from each of the loaders

cache_landmarks: bool = <dataclasses._MISSING_TYPE object>

True to cache landmarks from the other side for Warp to landmarks

flip: bool = <dataclasses._MISSING_TYPE object>

False to disable warping True to enable warping

folders: list[str] = <dataclasses._MISSING_TYPE object>

List of folders to be used as inputs to the model. Folders are provided in processing order (eg: [A, B, …])

lr_finder: bool = False

True to use the learning rate finder

snapshot_interval: int = -1

The number of iterations between snapshots

warp: bool = <dataclasses._MISSING_TYPE object>

False to disable warping True to enable warping