[Feature] Support timm backbones.

[RangiLyu]

Support load timm backbones

There are hundreds of pre-trained backbones in timm https://github.com/rwightman/pytorch-image-models

With this PR, you can directly use these backbones in NanoDet.

Usage

For example, to use mobilenet v3 in timm:

First install timm by running pip install timm

Then create a config like this

model:
  arch:
    name: NanoDetPlus
    detach_epoch: 10
    backbone:
      name: TIMMWrapper
      model_name: mobilenetv3_large_100
      features_only: True
      pretrained: True
      output_stride: 32
      out_indices: [2, 3, 4]
    fpn:
      name: GhostPAN
      in_channels: [40, 112, 960]
      ... ...

[codecov]

Codecov Report

Merging #399 (2e6d3e5) into main (b7600c8) will increase coverage by 0.08%.
The diff coverage is 76.92%.

@@            Coverage Diff             @@
##             main     #399      +/-   ##
==========================================
+ Coverage   74.36%   74.44%   +0.08%     
==========================================
  Files          66       67       +1     
  Lines        4486     4512      +26     
  Branches      756      760       +4     
==========================================
+ Hits         3336     3359      +23     
- Misses        970      972       +2     
- Partials      180      181       +1     

Flag	Coverage Δ
unittests	74.44% <76.92%> (+0.08%)	⬆️

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

Impacted Files	Coverage Δ
nanodet/model/backbone/__init__.py	93.10% <66.66%> (+0.79%)	⬆️
nanodet/model/backbone/timm_wrapper.py	78.26% <78.26%> (ø)
nanodet/data/transform/warp.py	85.63% <0.00%> (+1.14%)	⬆️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more