Skip to content

Latest commit

 

History

History
52 lines (39 loc) · 2.08 KB

config.md

File metadata and controls

52 lines (39 loc) · 2.08 KB

Docs for configs

Tips

How to confirm the parameter

  1. Confirm the model config

Please confirm the model you use. The model is in /projects/* and confirm what configs you should use. As an example, we use CenterPoint and this config file.

  1. Confirm the parameter to train

To confirm the parameters for training, please see like the parameters of config file.

  1. Confirm the dataset config

From the import file, confirm the dataset config in /autoware_ml/configs/detection3d/dataset/. As an example, please confirm like dataset config. An the yaml files are defined and it has the information of dataset.

How to manage the config files

The config file of MMlab libraries has high degree of freedom to set parameters. However, it is easy to lead technical debt because config files are scattered.

So we recommend following config directory structure.

- autoware_ml/
  - configs/
    - detection3d/
      - component.py
- projects/
  - TransFusion/
    - configs/
      - t4dataset/
        - 90m_768grid/
          - default.py
          - parameter_1.py
          - parameter_2.py
        - 50m_768grid/
          - default.py

Detection3d

Use for Tensorboard

  • Add backend for Tensorboard to config
vis_backends = [dict(type='LocalVisBackend'), dict(type='TensorboardVisBackend')]