[Refactory] Merge BEiT and ConvNext 's LR decay optimizer constructors

[linfangjian01]

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

As BEiT and ConvNext 's LR decay optimizer constructors have similar functions, the difference between them is how to get layer/stage id and we merge them in this pr.

Modification

1. Rename mmseg/core/utils/layer_decay_optimizer_constructor.py to mmseg/core/optimizers/layer_decay_optimizer_constructor.py
2. Remove mmseg/core/layer_decay_optimizer_constructor.py
3. Rename get_num_layer_layer_wise get_num_stage_stage_wist
4. Add a condition for BEiT in layer_wise
5. Add deprecated warning to LayerDecayOptimizerConstructor
6. Re-organize UTs based on modifications

[codecov]

Codecov Report

Merging #1438 (23b920d) into master (239049c) will increase coverage by 0.03%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #1438      +/-   ##
==========================================
+ Coverage   90.30%   90.34%   +0.03%     
==========================================
  Files         140      140              
  Lines        8345     8327      -18     
  Branches     1403     1400       -3     
==========================================
- Hits         7536     7523      -13     
+ Misses        571      569       -2     
+ Partials      238      235       -3     

Flag	Coverage Δ
unittests	90.34% <100.00%> (+0.03%)	⬆️

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

Impacted Files	Coverage Δ
mmseg/core/__init__.py	100.00% <100.00%> (ø)
mmseg/core/optimizers/__init__.py	100.00% <100.00%> (ø)
...re/optimizers/layer_decay_optimizer_constructor.py	96.11% <100.00%> (ø)
mmseg/core/utils/__init__.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 239049c...23b920d. Read the comment docs.

[Junjun2016]

Can we merge LayerDecayOptimizerConstructor and LearningRateDecayOptimizerConstructor to one class?

[linfangjian01]

Yep.

[MeowZheng]

must be add a deprecated warning for the deleted one

[MengzhangLI]

I suggest split LearningRateDecayOptimizerConstructor of ConvNeXt and BEiT only if you can ensure their training performances would be the same as before and log info does not leave out import message.

Because the diff shows the difference between original two LearningRateDecayOptimizerConstructor files not only lies in get_num_layer_for_vit function.

Btw, it is absolutely excellent if you can merge these two LearningRateDecayOptimizerConstructor into one file which means you would like to spend more efforts to check its detailed difference and performance.

[MengzhangLI]

I suggest we may remind users know what are these two optimizer construtors for. Because based on their name, users may not know relation between them. Related comments are flexible so you name it .