Match image by name (instead of id) on CVAT upload

[artemisart]

Motivation and context

Currently, the cvat annotation import (upload) matches annotations to frames by id, so this works only on very specific cases like backups for the same task. When moving annotations between tasks, if the order doesn't match exactly or if there are more frames, or (in our case) if we generate cvat annotations for pre-annotation, nothing matched and it's useless.
This PR just uses the same matching as for coco files, so it works both for old use-cases like backup (if the ID matched, then the filenames will match) but also pre-annotation and others.

How has this been tested?

Manually, uploaded some files.

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 have added tests to cover my changes
• 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)

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)

# Copyright (C) 2020 Intel Corporation
#
# SPDX-License-Identifier: MIT

[coveralls]

Pull Request Test Coverage Report for Build 6493

• 3 of 3 (100.0%) changed or added relevant lines in 1 file are covered.
• 699 unchanged lines in 28 files lost coverage.
• Overall coverage decreased (-0.2%) to 64.918%

---

Files with Coverage Reduction	New Missed Lines	%
cvat/apps/dataset_manager/formats/mask.py	1	89.29%
cvat/apps/dataset_manager/formats/mot.py	1	94.74%
datumaro/datumaro/plugins/datumaro_format/converter.py	2	98.16%
datumaro/datumaro/plugins/tf_detection_api_format/extractor.py	2	88.68%
datumaro/datumaro/plugins/coco_format/importer.py	3	79.49%
cvat/apps/dataset_manager/formats/pascal_voc.py	4	60.0%
datumaro/datumaro/cli/util/project.py	4	26.92%
datumaro/datumaro/plugins/yolo_format/converter.py	4	89.72%
datumaro/datumaro/util/init.py	4	80.85%
datumaro/datumaro/components/converter.py	5	88.46%

Totals
Change from base Build 6185:	-0.2%
Covered Lines:	11034
Relevant Lines:	16590

---

💛 - Coveralls

[zhiltsov-max]

I think you should use https://github.com/opencv/cvat/blob/develop/cvat/apps/dataset_manager/bindings.py#L571 here, which will fall back to ids and other things in the case image name did not match. This can happen, for example, in the case when the task was created from files on a shared storage, as image names in the task will include their absolute paths, in which case the new variant doesn't work. This problem would require analysis of the whole input file, but ids can simplify things.

Another simple variant - use the name, otherwise use the id and check if basenames match, if there is one provided.

[artemisart]

Indeed, I thought match_frame already matched on the basename and not the full path but not sure, and I didn't follow the updates of the importer.

[zhiltsov-max]

You can find an example of a DataseItem for these purposes here: https://github.com/opencv/cvat/blob/develop/cvat/apps/dataset_manager/formats/yolo.py#L40

I think, the approach in the last line above will just fit.

[azhavoro]

Hi, any updates?

[artemisart]

Done

[nmanovic]

@artemisart , thanks for the great contribution. I personally want to accept it. The only problem that it is necessary to fix a component from @zhiltsov-max. Are you going to do that?

[artemisart]

Yes, sorry I didn't have the time to update and test yet but it should be good now.

[nmanovic]

@artemisart , Thanks for the PR. Great contribution!