[Feature] Support ConvNeXt

[mzr1996]

Motivation

Support ConvNeXt and add converted checkpoints

Modification

As the title

Checklist

Before PR:

• Pre-commit or other linting tools are used to fix the potential lint issues.
• Bug fixes are fully covered by unit tests, the case that causes the bug should be added in the unit tests.
• The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
• The documentation has been modified accordingly, like docstring or example tutorials.

After PR:

• If the modification has potential influence on downstream or other related projects, this PR should be tested with those projects, like MMDet or MMSeg.
• CLA has been signed and all committers have signed the CLA in this PR.

[codecov]

Codecov Report

Merging #670 (9ef533b) into dev (8488a78) will increase coverage by 0.18%.
The diff coverage is 94.82%.

@@            Coverage Diff             @@
##              dev     #670      +/-   ##
==========================================
+ Coverage   82.81%   83.00%   +0.18%     
==========================================
  Files         123      124       +1     
  Lines        7293     7409     +116     
  Branches     1267     1289      +22     
==========================================
+ Hits         6040     6150     +110     
- Misses       1076     1077       +1     
- Partials      177      182       +5     

Flag	Coverage Δ
unittests	83.00% <94.82%> (+0.18%)	⬆️

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

Impacted Files	Coverage Δ
mmcls/models/backbones/convnext.py	94.78% <94.78%> (ø)
mmcls/models/backbones/__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 8488a78...9ef533b. Read the comment docs.

[Ezra-Yu]

channels_last make users confused, users may see it as letting input tensor as shape (N, H, W, C), channel-dim in the last, but actually, It has nothing to do with the input tensor in our implement

maybe replace it with use_linear(bool)=False like timm.

[Ezra-Yu]

LGTM.

[nijkah]

Hello, in the original paper and code, there is a mention about layer-wise decay.
Is it the future step to implement on mmcls?

[Ezra-Yu]

Hello, in the original paper and code, there is a mention about layer-wise decay. Is it the future step to implement on mmcls?

In the official repo, layer-wise decay is not used in the classification task but in downstream tasks like seg and det. So we have not added this feature in future plans.

And, this feature has been implemented in mmseg in open-mmlab/mmsegmentation#1216. If you want to use this in mmcls, you can transfer it from mmseg to mmcls.

[nijkah]

@Ezra-Yu Thank for your reply.
I was considering where I have to implement layer-wise decay when using convnext for mmdetection.