Fix potential bugs in accuracy.py

[Johnson-Wang]

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

There is a potential bug in accuracy.py, and This PR aims to fix this issue.
In line

mmsegmentation/mmseg/models/losses/accuracy.py

Line 48 in 618d3c3

correct = correct[:, target != ignore_index]

, correct = correct[:, target != ignore_index] is used to filter out the ignored targets. However, as the default value of ignore_index is None, this line would cause unwanted behavior. For example

>>> target = torch.ones(3)
>>> target != 255
tensor([True,  True, True])
>>> target != None
True
>>> target[target != 255].shape
torch.Size([3])
>>> target[target != None].shape
torch.Size([1, 3])

Therefore an unwanted additional dimension is created in the original implementation. Luckily, the correct tensor is finally flattened and the above behavior would not affect the final results. However, this is still a potential bug, and a quick fix would be better.

Modification

I used an if else condition to cater for the case of ignored_index is None.

Checklist

• Pre-commit or other linting tools are used to fix the potential lint issues.
• The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
• If the modification has potential influence on downstream projects, this PR should be tested with downstream projects, like MMDet or MMDet3D.
• The documentation has been modified accordingly, like docstring or example tutorials.

[CLAassistant]

All committers have signed the CLA.

[codecov]

Codecov Report

Merging #1496 (f2162bb) into master (618d3c3) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #1496   +/-   ##
=======================================
  Coverage   90.30%   90.30%           
=======================================
  Files         140      140           
  Lines        8335     8338    +3     
  Branches     1400     1402    +2     
=======================================
+ Hits         7527     7530    +3     
  Misses        570      570           
  Partials      238      238           

Flag	Coverage Δ
unittests	90.30% <100.00%> (+<0.01%)	⬆️

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

Impacted Files	Coverage Δ
mmseg/models/losses/accuracy.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 618d3c3...f2162bb. Read the comment docs.