Detection IoU

[twsl]

What does this PR do?

Fixes #659

Continues the work from #761 and adds IoU, GIoU, DIoU and CIoU

Things I did so far:

• Implementation follows naming convention
• The input and its validation is the same as for mAP so all detection metrics can be used in a MetricsCollection
• Threshold is set to None by default to enable proper progress logging, especially for IoUs with negative values
• Use .diag().mean() for single value calculation if all labels are equal, .mean() and all permutations otherwise
• Implemented additional metrics via inheritance
• Check for correct labels is supported and enabled by default
• IoUs are aggregated per image first and then averaged

Things to do:

• There is still no checking whether the labels are correct, how to handle them in calculation?
• Decide what to do with multiple boxes per ground truth, usually in object detection this is handled by NMS
• More test cases are required
• Currently IoU calculation takes all boxes and doesn't evaluate on a per image basis. Should it rather calculate on a per image basis or keep evaluation per box?
• is .diag().mean() the correct behaviour if the number if ground truth boxes != predicted boxes?

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 🙃

cc: @SkafteNicki @jscottcronin @tkupek @Borda

[SkafteNicki]

This is great work :) really looking forward to this being in TM
For now I have just taken an overall look, will try to deep dive into the code sometime during the next couple of days