What is the sigmas at configs/datsets/coco.py?

[seon-creator]

Hello, I'm searching sigmas concept for training custom datset.

The number of keypoints in my custom data set is 5.
And I set the sigma value for each keypoint to 0.025 each.
I set this value by referring to one of the sigma values of coco.py .
The inference result itself wasn't bad when I set it up like this, but can I expect a change in performance by adjusting this value?

I expect better accuracy so should I increase the sigma value or make it smaller?

According to the search, the smaller the value of sigma, the more accurate the inference of keypoint is, but I don't know if it's reliable information.

This is my custom coco dataset.py

dataset_info = dict(

    dataset_name='coco',    
    paper_info=dict(
        author='Lin, Tsung-Yi and Maire, Michael and '
        'Belongie, Serge and Hays, James and '
        'Perona, Pietro and Ramanan, Deva and '
        r'Doll{\'a}r, Piotr and Zitnick, C Lawrence',
        title='Microsoft coco: Common objects in context',
        container='European conference on computer vision',
        year='2014',
        homepage='http://cocodataset.org/',
    ),
    keypoint_info={
        0:
        dict(
            name='1',
            id=0,
            color=[255,255,255],
            type='upper',
            swap='1'),
        1:
        dict(
            name='2',
            id=1,
            color=[255,255,255],
            type='upper',
            swap='2'),
        2:
        dict(
            name='3',
            id=2,
            color=[255,255,255],
            type='upper',
            swap='3'),
        3:
        dict(
            name='4',
            id=3,
            color=[255,255,255],
            type='upper',
            swap='4'),
        4:
        dict(
            name='5',
            id=4,
            color=[255,255,255],
            type='upper',
            swap='5')

    },
    skeleton_info={     
                       
        0:
        dict(link=('1', '1'), id=0, color=[255, 0, 0]),
        1:
        dict(link=('2', '2'), id=1, color=[255, 0, 0]),
        2:
        dict(link=('3', '3'), id=2, color=[255, 0, 0]),
        3:
        dict(link=('4', '4'), id=3, color=[255, 0, 0]),
        4:
        dict(link=('5', '5'), id=4, color=[255, 0, 0]),
    },
    joint_weights=[
        1., 1., 1., 1., 1.
    ],
    sigmas=[
        0.025, 0.025, 0.025, 0.025, 0.025
    ]
)

[Ben-Louis]

Please refer to https://cocodataset.org/#keypoints-eval.

[seon-creator]

Thank you :)
Is there any method(tools)in mmpose for calculate sigmas for custom dataset?
If not, should I have to change sigmas and check the performance by experiment?