Provide custom task name for CVAT

[Rusteam]

What changes are proposed in this pull request?

Closes #1753

1. Custom task name can be passed for labelling data in CVAT such as dataset.annotate("anno_key", backend="cvat", task_name="Custom task name", ...)
2. Default task name for CVAT to include an annotation key such as FiftyOne_{dataset_name}_{annotation_key}

How is this patch tested? If it is not, please explain why.

added few lines to existing cvat unit-tests.

Release Notes

Is this a user-facing change that should be mentioned in the release notes?

• No. You can skip the rest of this section.
• Yes. Give a description of this change to be included in the release
  notes for FiftyOne users.

A custom task name can be provided with CVAT annotation backend. Default CVAT task name now includes an annotation key.

What areas of FiftyOne does this PR affect?

• App: FiftyOne application changes
• Build: Build and test infrastructure changes
• Core: Core fiftyone Python library changes
• Documentation: FiftyOne documentation changes
• Other

[Rusteam]

@ehofesmann just have a couple of notes about this implementation:

1. I had to make a slight workaround to get an annotation key from annotation_backend.upload_samples() in order to get the latest annotation run key without changing the main api for all backends.
2. I was unable to build docs to check that the document update generated nicely because of an error in bash generate_docs.bash. Should I file an issue? Here is the traceback:

Generating model zoo listing page
Traceback (most recent call last):
  File "/Users/rustem.galiullin/Projects/fiftyone/docs/scripts/make_model_zoo_docs.py", line 17, in <module>
    import fiftyone.zoo as foz
  File "/Users/rustem.galiullin/Projects/fiftyone/fiftyone/zoo/__init__.py", line 9, in <module>
    from .datasets import *
  File "/Users/rustem.galiullin/Projects/fiftyone/fiftyone/zoo/datasets/__init__.py", line 20, in <module>
    import fiftyone.utils.data as foud
  File "/Users/rustem.galiullin/Projects/fiftyone/fiftyone/utils/data/__init__.py", line 10, in <module>
    from .base import *
  File "/Users/rustem.galiullin/Projects/fiftyone/fiftyone/utils/data/base.py", line 20, in <module>
    import fiftyone.core.fields as fof
  File "/Users/rustem.galiullin/Projects/fiftyone/fiftyone/core/fields.py", line 21, in <module>
    import fiftyone.core.odm as foo
  File "/Users/rustem.galiullin/Projects/fiftyone/fiftyone/core/odm/__init__.py", line 10, in <module>
    from .database import (
  File "/Users/rustem.galiullin/Projects/fiftyone/fiftyone/core/odm/database.py", line 64, in <module>
    class DatabaseConfigDocument(Document):
  File "/Users/rustem.galiullin/Projects/fiftyone/fiftyone/core/odm/database.py", line 70, in DatabaseConfigDocument
    version = fof.StringField()
AttributeError: partially initialized module 'fiftyone.core.fields' has no attribute 'StringField' (most likely due to a circular import)

[brimoor]

This is a clever workaround to get the anno_key, but now that config.task_name is supported, I think it would be best to just rely on the user putting their anno_key in the task name if that's what they want.

[brimoor]

It is unfortunate that we didn't include anno_key in either the upload_annotations() method or in the AnnotationBackendConfig, because it's definitely reasonable to want to use it to generate default values like this 🤦

[Rusteam]

I guess it's enough for me that I can provide a custom task name to differentiate between tasks for the same dataset.

Do you think that will require a lot of/breaking changes to pass anno_key to upload_annotations?

[brimoor]

In cases where there are more than one task in the run, what do you think about treating tasks_name as a base name and appending 1, 2, 3, etc to it?

[Rusteam]

Would not that require getting the count of annotation keys as well?

[brimoor]

No I'm suggesting adding a separate enhancement to this PR: a single annotation run may create multiple tasks (eg 1 per video for video collections, or if the task_size parameter is provided for image collections). I'm suggesting that you update the logic so that the tasks within the run are named task_name1, task_name2`, etc.

[Rusteam]

Oh yeah that can be done.

How about the default task name? Do we include an annotation key or not?

[brimoor]

How about the default task name? Do we include an annotation key or not?

I don't think we should include annotation key in the default task name because there's no official way to get that information into upload_samples() right now without making a breaking change to the annotation API.

[brimoor]

LGTM, thanks for the contribution!