Add ExtractedMask and update importers who can use it to use it

[vinnamkim]

Summary

• Ticket no. 138711
• Add ExtractedMask which is a dedicated mask annotation class to use a single index mask source. The index mask is a integer 2D array and its pixel can indicate a label id (class) or instance id.
• The advantage of introducing this class is that we can minimize a memory burden by sharing a single source explicitly. In addition, we can use a computational burden for the following scenario:

	Create a semantic segmentation map by combining Datumaro Mask annotations
Setup	Assume there is a Datumaro dataset item and it has multiple Datumaro Mask. We can merge them into one index mask to construct a semantic segmentation map.
Before	1) For each Datumaro Mask annotation, create a binary masks, 2) Merge binary masks into one index mask
After	1) Check all Mask annotations are ExtractedMask and it share the same index mask, 2) Just extract the source index mask

It can save huge computations required for 2D array manipulations: creating a binary mask and merging binary masks.

How to test

Checklist

• I have added unit tests to cover my changes.​
• I have added integration tests to cover my changes.​
• I have added the description of my changes into CHANGELOG.​
• I have updated the documentation accordingly

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) 2024 Intel Corporation
#
# SPDX-License-Identifier: MIT

[codecov]

Codecov Report

Attention: Patch coverage is 98.33333% with 1 lines in your changes are missing coverage. Please review.

Project coverage is 80.84%. Comparing base (44cc56a) to head (af03f29).
Report is 47 commits behind head on develop.

Files	Patch %	Lines
src/datumaro/plugins/data_formats/voc/base.py	75.00%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #1480      +/-   ##
===========================================
- Coverage    80.85%   80.84%   -0.02%     
===========================================
  Files          271      273       +2     
  Lines        30689    31402     +713     
  Branches      6197     6316     +119     
===========================================
+ Hits         24815    25386     +571     
- Misses        4489     4592     +103     
- Partials      1385     1424      +39     

Flag	Coverage Δ
ubuntu-20.04_Python-3.10	80.82% <98.33%> (-0.03%)	⬇️
windows-2022_Python-3.10	80.81% <98.33%> (-0.02%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[wonjuleee]

What is the difference between index and label here?

[vinnamkim]

As described here

datumaro/src/datumaro/components/annotation.py

Lines 513 to 515 in e2e945e

	Attributes:
	index_mask: Integer 2D Numpy array. Its pixel can indicate a label id (class) or an instance id.
	index: Integer value to extract a binary mask from the given index mask.

index can denote an instance id or a label id. It depends on the dataset. Anyway, I enhanced the example docstring in af03f29 to make it more descriptive.

[wonjuleee]

I am still confused. The updated example have the same index and label. In my understanding, mask1 is a binary mask with 0 and 1, where all 1s indicate the index indeed. This looks fine. But index might be label for all cases. Is there any case that index is not equal to label?

[vinnamkim]

mask is a binary mask and its 1s actually indicate label=1 (not index=1) in this example. index is just used to extract a binary mask from index_mask. It is possible that index can be different with label by dataset transformation. For example, if the label remapping transform merges label=1, index=1 (1: apple) and label=2, index=2 (2: orange) to label=3 (3: fruits). Then, apple masks will have index=1, label=3 and orange masks will have index=2, label=3.