ParticleDetection.modelling.configs

Collection of helper functions to be used with CfgNode configuration objects for a Detectron2 network. Additionally, defines a ported version of a previously used R-CNN from an older implementation.

Author: Adrian Niemann (adrian.niemann@ovgu.de)

Date: 31.10.2022

PORTED_AUGMENTATIONS = [detectron2.data.transforms.AugmentationList]

List of augmentations used during training of a rod detection network.

get_epochs(cfg: detectron2.config.config.CfgNode, image_count: int) → float[source]

Computes the achieved number of epochs with given settings and data.

Parameters

  • cfg (CfgNode) – Configuration of a network to be trained with Detectron2. Must contain at least the keys SOLVER.IMS_PER_BATCH and SOLVER.MAX_ITER.

  • image_count (int) – Number of images the network will be trained on.

Returns

float

get_iters(cfg: detectron2.config.config.CfgNode, image_count: int, desired_epochs: int) → int[source]

Computes the necessary iterations to achieve a given number of epochs.

Parameters

  • cfg (CfgNode) – Configuration of a network to be trained with Detectron2. Must contain at least the key SOLVER.IMS_PER_BATCH.

  • image_count (int) – Number of images the network will be trained on.

  • desired_epochs (int) – Number of epochs the network shall be trained using a dataset with ìmage_count number of images.

Returns

int

old_ported_config(dataset: Optional[DataSet] = None, test_dataset: Optional[DataSet] = None) → detectron2.config.config.CfgNode[source]

Creates a configuration resembling one previously used with an older implementation of a R-CNN.

Parameters

  • dataset (DataSet) –

    Dataset intended for training the network.

    Default is None.

  • test_dataset (DataSet) – Dataset for testing the network’s performance during training. Default is None.

Returns

CfgNode

run_test_config(dataset: DataSet) → detectron2.config.config.CfgNode[source]

Creates a configuration with only few iterations for testing new code.

write_configs(cfg: detectron2.config.config.CfgNode, directory: str, augmentations: Optional[List[detectron2.data.transforms.Augmentation]] = None) → None[source]

Write network configurations to a target directory.

Writes a config.yaml file from the configuration and possibly an augmentations.pkl file.

Parameters

  • cfg (CfgNode) – Configuration for a network handled with Detectron2.

  • directory (str) – Directory the configuration shall be written to.

  • augmentations (List[Augmentation]) – List of image augmentations that shall be saved alongside the network configuration. Default is None.