fix ignore_index in the computation of IoU

[CSautier]

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?

What does this PR do?

Fixes #312

PR review

The way it does is to remove the line and columns of the ignore_index in the confmat. It should work whenever ignore_index is in the correct range but could be tested further.

There are certainly cleaner ways to do this.

[codecov]

Codecov Report

Merging #328 (32389fb) into master (85ebc3b) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #328   +/-   ##
=======================================
  Coverage   96.08%   96.08%           
=======================================
  Files         124      124           
  Lines        3980     3982    +2     
=======================================
+ Hits         3824     3826    +2     
  Misses        156      156           

Flag	Coverage Δ
Linux	75.69% <50.00%> (-0.02%)	⬇️
Windows	75.69% <50.00%> (-0.02%)	⬇️
cpu	96.08% <100.00%> (+<0.01%)	⬆️
macOS	96.08% <100.00%> (+<0.01%)	⬆️
pytest	96.08% <100.00%> (+<0.01%)	⬆️
python3.6	95.30% <100.00%> (+<0.01%)	⬆️
python3.8	96.08% <100.00%> (+<0.01%)	⬆️
python3.9	95.98% <100.00%> (+<0.01%)	⬆️
torch1.3.1	95.30% <100.00%> (+<0.01%)	⬆️
torch1.4.0	95.37% <100.00%> (+<0.01%)	⬆️
torch1.9.0	95.98% <100.00%> (+<0.01%)	⬆️

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

Impacted Files	Coverage Δ
torchmetrics/functional/classification/iou.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 85ebc3b...32389fb. Read the comment docs.

[Borda]

nice, can we add a test to verify it works as intended now?

[Borda]

@CSautier mind check the failing tests?

[Borda]

@CSautier how is it going? 🐰

[CSautier]

Sorry, it slipped out of my mind.
I verified the tests, and I believe test_iou(True, 'none', 0, torch.Tensor([0.5, 0.5])) does not make sense. I think the result should indeed be (2/3, 1/2) instead of (1/2, 1/2), which would mean that the current version passing this test is wrong. Similarly test_iou_ignore_index([0, 1, 1, 2, 2], [0, 1, 2, 2, 2], 2, 3, 'none', [1, 1 / 2]) should indeed return (1,1) as my version does and not (1, 1/2) as the test expects.

However there is one correct tests that my computation does not pass
test_iou_ignore_index([0, 1, 1, 2, 2], [0, 1, 2, 2, 2], 1, 3, 'none', [1, 2 / 3])
and it shows a problem in how I remove the ignore_index if it is predicted.

I'm working on it now, and will change my PR if I find a solution.
If you could also verify that I'm correct and test_iou(True, 'none', 0, torch.Tensor([0.5, 0.5])) should indeed return (2/3, 1/2) and test_iou_ignore_index([0, 1, 1, 2, 2], [0, 1, 2, 2, 2], 2, 3, 'none', [1, 1 / 2]) should in fact return (1, 1) that would be great.

[pep8speaks]

Hello @CSautier! Thanks for updating this PR.

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2021-07-29 16:03:10 UTC

[CSautier]

I've updated the PR, and included the two modifications I mentioned earlier in the tests.

[SkafteNicki]

LGTM :]