Configuration files

Configuration files#

Configuration files in the config folder define all parameters for training, validation, and testing. Each config is written in YAML format and is used to initialize model and dataset classes using two keys:

  • class_path — specifies which class will be used (for example, for the model or dataset).

  • init_args — sets the arguments that will be passed to the constructor of the selected class.

Parameters for the trainer section correspond to the arguments of the Lightning Trainer class. See the official documentation for all available options: Trainer.

You can use the provided configs as templates or create your own.

Config file structure#

seed_everything: 42

model:
  class_path: model.Yolo
  init_args:
    model: l
    num_classes: 2
    # ... other model parameters ...
    plotter:
      class_path: utils.Plotter
      init_args:
        threshold: 0.0
        show_video: true
        # ... other plotter parameters ...

data:
  class_path: utils.PropheseeDataModule
  init_args:
    root: ./data/gen1
    batch_size: 8
    # ... other data parameters ...

trainer:
  accelerator: gpu
  strategy: auto
  devices: -1
  # ... other trainer parameters ...

# Optionally, you can add callbacks, logger, etc.

Available values for class_path#

For models:

For datasets:

For plotter:

Parameter descriptions#

For a full list and description of available parameters for each class, see the API section:

Note

The classes model.Yolo and model.MultimodalYolo both inherit from model.Detector. All initialization parameters of model.Detector are also available for these models.

See the API section for constructor signatures and parameter documentation.

Contents