Fix mIoU calculatiton range

[clownrat6]

1. Fix the bug about torch.histc function. When setting arg:max same as arg:bins for torch.histc, some channel of mIoU may be nan.
2. Modify tests/test_metrics.py. Add some abnormal num_classes test cases.

Pascal Context Dataset

Single-scale, No flip, mIoU test result:

model	num classes	Lr schd	after fix	before fix	readme
HRNetV2-W48(new)	59	40000	50.37	50.13	None
HRNetV2-w48(new)	59	80000	51.12	50.85	None
HRNetV2-W48(new)	60	40000	45.55	45.55	45.14
HRNetV2-W48(old)	60	40000	45.14	45.14	45.14

before fix:

• keyboard=nan

Cityscapes Dataset

Single-scale, No flip, mIoU test result:

model	Crop Size	Lr schd	after fix	before fix	readme
HRNetV2-W48(old)	512x1024	40000	78.52	78.52	78.48
HRNetV2-W48(old)	512x1024	80000	79.98	79.98	79.93

ADE20K Dataset

Single-scale, No flip, mIoU test result:

model	Crop Size	Lr schd	after fix	before fix	readme
HRNetV2-W48(old)	512x512	80000	41.9	41.76	41.9
HRNetV2-W48(old)	512x512	160000	42.02	41.82	42.02

before fix:

• fireplace=nan;
• blind=nan;
• television receiver=nan;
• botle=nan;
• cradle=nan;
• animal=nan;

[codecov]

Codecov Report

Merging #471 (031883a) into master (e0e985f) will decrease coverage by 0.07%.
The diff coverage is 16.66%.

❗ Current head 031883a differs from pull request most recent head fb73614. Consider uploading reports for the commit fb73614 to get more accurate results

@@            Coverage Diff             @@
##           master     #471      +/-   ##
==========================================
- Coverage   86.58%   86.50%   -0.08%     
==========================================
  Files          97       97              
  Lines        4964     4966       +2     
  Branches      807      806       -1     
==========================================
- Hits         4298     4296       -2     
- Misses        514      517       +3     
- Partials      152      153       +1     

Flag	Coverage Δ
unittests	86.50% <16.66%> (-0.08%)	⬇️

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

Impacted Files	Coverage Δ
mmseg/apis/inference.py	63.79% <0.00%> (-2.28%)	⬇️
mmseg/core/evaluation/metrics.py	86.56% <ø> (ø)
mmseg/ops/wrappers.py	80.76% <ø> (-1.99%)	⬇️
mmseg/models/segmentors/encoder_decoder.py	90.84% <33.33%> (-1.16%)	⬇️

---

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 e0e985f...fb73614. Read the comment docs.

[xvjiarui]

Could we add some test cases for this?

[xvjiarui]

For example, we may construct some test cases that fail in the previous version.

[clownrat6]

We can use trained model file to test on both two version IoU calculation function.
The trained model file BaiduNetDisk(Extract Code: 0l69)

[xvjiarui]

How does this PR affect mIoU of cityscapes and ADE20k?

[clownrat6]

How does this PR affect mIoU of cityscapes and ADE20k?

mIoU of cityscapes has a little bit wave and doesn't contain nan value. But mIoU of ADE20K has six nan classes.

[OpenMMLab-Coodinator]

Hi @Sennnnn ！First of all, we want to express our gratitude for your significant PR in this project. Your contribution is highly appreciated, and we are grateful for your efforts in helping improve this open-source project during your personal time. We believe that many developers will benefit from your PR.

We would also like to invite you to join our Special Interest Group (SIG) private channel on Discord, where you can share your experiences, ideas, and build connections with like-minded peers. To join the SIG channel, simply message moderator— OpenMMLab on Discord or briefly share your open-source contributions in the #introductions channel and we will assist you. Look forward to seeing you there! Join us ：https://discord.gg/raweFPmdzG

If you are Chinese or have WeChat，welcome to join our community on WeChat. You can add our assistant ：openmmlabwx. Please add "mmsig + Github ID" as a remark when adding friends：）
Thank you again for your contribution❤