[Feature] timm backbones wrapper

[amirassov]

Motivation

TIMM (https://github.com/rwightman/pytorch-image-models) contains a lot of backbones, which are not available in this repository. But mmclassification code is more flexible for training and adding new components. It would be convenient to have wrapper here to use TIMM backbones.

Modification

This PR adds TIMMBackbone wrapper to use timm backbones.

Use cases (Optional)

To use Efficientnetv2-s from timm with pretrained ImageNet-21k weights you need to use the following backbone config:

backbone=dict(
    type="TIMMBackbone",
    pretrained=True,
    model_name="tf_efficientnetv2_s_in21k",
    init_cfg=None
)

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.

[CLAassistant]

All committers have signed the CLA.

[mzr1996]

Thanks for your contribution!
We will review this PR recently.
Usually, our backbones need to support downstream repos, so we can't use timm backbone directly. But if someone just want to use our platform to train third-party backbones, this function will be helpful.

[mzr1996]

Hello, here is a problem that breaks the Python3.6 CI.

Python 3.6.13 |Anaconda, Inc.| (default, Jun  4 2021, 14:25:59) 
[GCC 7.5.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import timm
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/xxx/miniconda3/envs/py36/lib/python3.6/site-packages/timm/__init__.py", line 2, in <module>
    from .models import create_model, list_models, is_model, list_modules, model_entrypoint, \
  File "/xxx/miniconda3/envs/py36/lib/python3.6/site-packages/timm/models/__init__.py", line 1, in <module>
    from .byoanet import *
  File "/xxx/miniconda3/envs/py36/lib/python3.6/site-packages/timm/models/byoanet.py", line 16, in <module>
    from .byobnet import ByoBlockCfg, ByoModelCfg, ByobNet, interleave_blocks
  File "/xxx/miniconda3/envs/py36/lib/python3.6/site-packages/timm/models/byobnet.py", line 28, in <module>
    from dataclasses import dataclass, field, replace
ModuleNotFoundError: No module named 'dataclasses'

timm uses dataclasses which is added in Python 3.7.

[codecov]

Codecov Report

Merging #427 (59cd4ff) into master (729c6c1) will increase coverage by 0.01%.
The diff coverage is 81.81%.

@@            Coverage Diff             @@
##           master     #427      +/-   ##
==========================================
+ Coverage   77.02%   77.04%   +0.01%     
==========================================
  Files          99      100       +1     
  Lines        5280     5302      +22     
  Branches      866      869       +3     
==========================================
+ Hits         4067     4085      +18     
- Misses       1093     1096       +3     
- Partials      120      121       +1     

Flag	Coverage Δ
unittests	77.04% <81.81%> (+0.01%)	⬆️

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

Impacted Files	Coverage Δ
mmcls/models/backbones/timm_backbone.py	80.95% <80.95%> (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 729c6c1...59cd4ff. Read the comment docs.

[mzr1996]

LGTM