Support automatic tag annotation

[subdavis]

Motivation and context

resolves #3358

How has this been tested?

I have a full-frame classifier nuclio lambda and needed a way to save these tags.

• I ran the code in development using docker-compose -f docker-compose.yml -f components/serverless/docker-compose.serverless.yml up -d
• Invoke my nuclio classifier from the task summary page (full video) and from the annotator (single frame)
• See new tags appear in the UI.

I did see that the author of #3358 has included confidence values, which I do not believe are supported by tags. If I should support this somewhere, please LMK.

Checklist

• I submit my changes into the develop branch
• I have added description of my changes into CHANGELOG file
• I have updated the documentation accordingly I do not see any documentation about what response types are supported for nuclio functions, so I don't think there's any documentation to update. New documentation could be added about the general capabilities of this feature, but it seems out of scope for this PR.
• I have added tests to cover my changes I do not see tests related to this section of the code. If they exist, please LMK and I'll add coverage.
• I have linked related issues (read github docs)
• I have increased versions of npm packages if it is necessary (cvat-canvas,
  cvat-core, cvat-data and cvat-ui) Not applicable

License

• I submit my code changes under the same MIT License that covers the project.
  Feel free to contact the maintainers if that's a concern.
• I have updated the license header for each file (see an example below) No new files added

# Copyright (C) 2021 Intel Corporation
#
# SPDX-License-Identifier: MIT

[nmanovic]

@subdavis , Could you please update the changelog? After we released 1.7.0 it isn't valid anymore.

[subdavis]

Yep, I can do that.

[nmanovic]

@subdavis , Could you please contribute any serverless function for classification? Thus we can test the PR.

[subdavis]

I don't think the introduction of a new serverless function is appropriately within scope for #3358

I actually don't even have one I can share: the code I have that generates full-frame detections is closed source. I can provide some instructions about how to temporarily modify an existing function, but I don't really have time to write a new one. I'm not an ML developer.

The problem here seems to be one of documentation: the only way to understand the API between CVAT and Nuclio functions is to read this views.py file, so I think documentation would be ideal.

[nmanovic]

@subdavis , it will be great to hear how in your company (kitware?) you are using CVAT to annotate data.

[subdavis]

We use CVAT occasionally at Kitware, but we're a consultancy and this PR was made for one of our customers who uses CVAT heavily in their own annotation processes.

[nmanovic]

@subdavis , thanks for sharing the information.

[nmanovic]

@subdavis , please let me know if you have time to answer on my comments?

[subdavis]

Using any of the existing functions, you should be able to test by swapping main with this. You'll need a dataset with an existing label called "example"

import json

def init_context(context):
    pass

def handler(context, event):
    results = [{
        'type': 'tag',
        'label': 'example',
    }]
    return context.Response(body=json.dumps(results), headers={},
        content_type='application/json', status_code=200)

[nmanovic]

LGTM