tlc.integration.detectron2.detectron_metrics_collection_hook#

Hook that collects detectron2 standard metrics.

Module Contents#

Classes#

Class

Description

DetectronMetricsCollectionHook

Hook that collects detectron2 standard metrics during training.

API#

class tlc.integration.detectron2.detectron_metrics_collection_hook.DetectronMetricsCollectionHook(run_url: tlc.core.objects.mutable_objects.run.Run | tlc.core.url.Url | str | None = None, cfg: detectron2.config.CfgNode | None = None, collection_start_iteration: int = 0, collection_frequency: int = -1, collect_metrics_before_train: bool = False)#

Bases: detectron2.engine.hooks.HookBase

Hook that collects detectron2 standard metrics during training.

Collect any metrics that are available in the detectron2 event storage and write them as a single table to the run.

Parameters:

  • run_url – A tlc.Run object, or the URL of the Run to use. Will default to using the current active Run.

  • cfg – The detectron2 config. If None, the config will be read from the trainer.

  • collection_start_iteration – The iteration to start collecting metrics on.

  • collection_frequency – The frequency with which to collect metrics.

  • collect_metrics_before_train – Indicates whether metrics should be collected at the beginning of training. Note: This option is deprecated and will be removed in future releases. The hook is intended primarily for collecting metrics during the training process, so using this argument is not recommended.

before_train() → None#

Checks that the detectron config includes the values required to collect metrics, collects metrics if required.

after_train() → None#

Writes the metrics table to the run.

after_step() → None#

Collects metrics if required.