Updated the mmdet2torchserve function:

[Danil328]

Motivation

I use mmdetection in my personal projects.

Modification

Updated the mmdet2torchserve function according to Torch Model archiver for TorchServe:

• added the ability to use a custom handler;
• added the ability to use extra files;
• fixed a bug when the mmdetection python config was supplied to the config_file input, as a result of which the code failed with a parsing YAML error.

[CLAassistant]

All committers have signed the CLA.

[Danil328]

The test "maximum_version_cpu" failed.
I didn't make any changes that could affect it.

FAILED tests/test_datasets/test_transforms/test_transforms.py::TestCorrupt::test_transform - TypeError: gaussian() got an unexpected keyword argument 'multichannel'

The problem is in the call to the gaussian function in the imagecorruptions library.
I discovered that multichannel is used in the scikit-image library up to version 0.19, starting from version 20 this parameter is not present.

Automatic detection of the color channel based on the old deprecated `multichannel=None` was broken in version 0.19. In 0.20 this behavior is fixed.

https://github.com/scikit-image/scikit-image/blob/441fe68b95a86d4ae2a351311a0c39a4232b6521/skimage/_shared/filters.py#L15

[Danil328]

@jbwang1997
Сan you please look at the pull request?

[Danil328]

Сan you review my pull request, please?
@jbwang1997 @Keiku @anurag1paul