[CodeCamp2023-327]Control whether methods of the Visualizer are executed only in the main process through a parameter

[DeliMm]

Motivation

Control whether methods of the Visualizer are executed only in the main process through a parameter. See more details in #1081

Modification

To implement these functions, I add a parameter img_only_master to visualizer through vis_backend. And I modified these files :

• modified mmengine/dist/utils.py
• modified mmengine/visualization/vis_backend.py
• modified mmengine/visualization/visualizer.py

During runtime, it is determined whether to retain only the images in the main GPU by checking if img_only_master is passed and examining its corresponding value. The following are modified master_only functions:

def master_only(func: Callable) -> Callable:
    """Decorate those methods which should be executed in master process.

    Args:
        func (callable): Function to be decorated.

    Returns:
        callable: Return decorated function.
    """

    @functools.wraps(func)
    def wrapper(*args, **kwargs):
        if ('img_only_master' in kwargs
                and kwargs['img_only_master'] is False) or is_main_process():
            return func(*args, **kwargs)	           
    return wrapper

When we build the runner, we can use this method to pass in img_only_master.

visualizer=dict(
            type='Visualizer',
            vis_backends=[
                dict(
                    type='LocalVisBackend',
                    save_dir='./vis',
                    img_only_master=False)
            ]),

When we use add_image in the training process, we also need to pass the extra parameter: img_only_master=runner.visualizer.img_only_master

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
3. If the modification has potential influence on downstream projects, this PR should be tested with downstream projects, like MMDet or MMCls.
4. The documentation has been modified accordingly, like docstring or example tutorials.

[CLAassistant]

All committers have signed the CLA.