Detection: panoptic quality

[niberger]

What does this PR do?

Add the panoptic quality metric, as defined in the reference paper.
This implementation differ from the reference implementation of the COCO dataset on the following points:

• The categories have to be declared at metric initialization.
• Any undefined category will be considered as an unlabelled (void) region.
• The input format for panoptic images are Tensor of shape (height, width, 2), where the first number of each pixel is the category id and the second number is the instance id.
• No support for group of instances (crowds in COCO).

Fixes #50
closes #1435

Before submitting

• Was this discussed/approved via a Github issue? (no need for typos and docs improvements)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure to update the docs?
• Did you write any new necessary tests?

PR review

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.

Did you have fun?

Make sure you had fun coding 🙃

[codecov]

Codecov Report

Merging #929 (781fe83) into master (61c2385) will decrease coverage by 51%.
The diff coverage is 93%.

Additional details and impacted files

@@           Coverage Diff            @@
##           master    #929     +/-   ##
========================================
- Coverage      88%     37%    -51%     
========================================
  Files         213     216      +3     
  Lines       10987   11139    +152     
========================================
- Hits         9660    4146   -5514     
- Misses       1327    6993   +5666     

[SkafteNicki]

few comments

[marcocaccin]

Hi @Borda, I just reached out on PTL's Slack to offer help if it's useful to have another pair of hands to bring this PR to the finish line ;)
Quick question: Metric.update() is expected to work on inputs with a batch dimension, right? This implementation works for individual samples only.

[justusschock]

Hey @marcocaccin

yes, metrics should always work with batches.
Help would be very welcome.

[justusschock]

@marcocaccin we merged it for now. Feel free to do another PR to add batched-support and also maybe more tests :)

[ChristophReich1996]

Hi @niberger, I just wanted to ask if you have verified that your PQ implementation matches the results of the COCO PQ implementation? Thanks :)