Fix LFW format

[sizov-kirill]

Summary

Resolves cvat-ai/cvat#5100

How to test

Checklist

• I submit my changes into the develop branch
• I have added description of my changes into CHANGELOG
• I have updated the documentation accordingly
• I have added tests to cover my changes
• I have linked related issues

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) 2021 Intel Corporation
#
# SPDX-License-Identifier: MIT

[yasakova-anastasia]

We don't get any annotations when we import if we only export points. This is true? It's correct in terms of format, but I'm not sure that it's good for CVAT. Maybe generalize this format? We can save both points and labels if we only have points. That is, when importing such a dataset, we will add extra labels, but the points will not disappear.

[sizov-kirill]

We don't get any annotations when we import if we only export points. This is true? It's correct in terms of format, but I'm not sure that it's good for CVAT. Maybe generalize this format? We can save both points and labels if we only have points. That is, when importing such a dataset, we will add extra labels, but the points will not disappear.

Yes, it's true. If we have only points annotations exported dataset in LFW format will be empty.

I didn't choose the solution you suggested for the following reasons:

1. If user annotated one image with few Point objects that have different labels it's not clear how to represent such information in LFW format, because in LFW landmarks are described with the following format <img> <points> so we don't have any opportunity to set corresponding between specific point and specific label
2. If during export we try to add Tags that aren't represented in original annotations it's not clear which labels we should to add. Probably if we have points on image that all have one label it's not problem. But if we have points with different label on the same image it's not clear which label we should chose for Label object. Of course we can add multiply Labels for one image , but format doesn't assume such usage.

But I understand that this is not an obvious question and it is difficult to find only one "correct" solution.

@zhiltsov-max How do you think?

[yasakova-anastasia]

Could you please update the Changelog?

I also think the documentation needs to be updated.

[zhiltsov-max]

I don't think this behavior in the test is right. From the format viewpoint, the input annotations are incomplete, and it doesn't seem we can adequately support such cases in the format, so the best approach would be either to complete the missing annotations, if possible, or raise a clear error message otherwise.

My considerations on this specific case with CVAT export are here. Let's try to make the format clearly usable from CVAT, Datumaro seem to be correct at this point.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarCloud