fix res2net+dcn test bug

[yuzhj]

I rewrite init_weights in the res2net.py to fix the test bug(AttributeError: 'Bottle2neck' object has no attribute 'conv2')

[codecov]

Codecov Report

Merging #3714 into master will increase coverage by 0.27%.
The diff coverage is 55.55%.

@@            Coverage Diff             @@
##           master    #3714      +/-   ##
==========================================
+ Coverage   60.83%   61.10%   +0.27%     
==========================================
  Files         216      216              
  Lines       15316    15340      +24     
  Branches     2607     2620      +13     
==========================================
+ Hits         9317     9374      +57     
+ Misses       5535     5497      -38     
- Partials      464      469       +5     

Flag	Coverage Δ
#unittests	61.10% <55.55%> (+0.27%)	⬆️

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

Impacted Files	Coverage Δ
mmdet/models/backbones/res2net.py	82.67% <52.00%> (-7.62%)	⬇️
mmdet/core/anchor/anchor_generator.py	77.53% <100.00%> (ø)
mmdet/models/roi_heads/mask_heads/fcn_mask_head.py	74.67% <0.00%> (+1.29%)	⬆️
mmdet/models/dense_heads/corner_head.py	76.22% <0.00%> (+1.91%)	⬆️
mmdet/models/detectors/cornernet.py	100.00% <0.00%> (+5.12%)	⬆️
mmdet/models/roi_heads/mask_heads/maskiou_head.py	97.87% <0.00%> (+5.31%)	⬆️
mmdet/models/roi_heads/test_mixins.py	60.17% <0.00%> (+9.73%)	⬆️
mmdet/models/roi_heads/mask_scoring_roi_head.py	91.07% <0.00%> (+32.14%)	⬆️

---

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 7cf344e...6fb9f1a. Read the comment docs.

[yuzhj]

the format problem is so difficult!!!!!!!!!!

[xvjiarui]

We may add comments here.
Like dcn in Res2Net bottleneck is in ModuleList.

[yuzhj]

We may add comments here.
Like dcn in Res2Net bottleneck is in ModuleList.

ok, i will add comments

[xvjiarui]

the format problem is so difficult!!!!!!!!!!

Hi @yuzhj
Thanks for your contribution. You may check contribute guide for reference.