[Feaute]support spconv2.0

[VVsssssk]

Motivation

Support spconv2.0 in mmdet3d dev branch. If the user installed spconv2.0, the code would use spconv2.0 first. And if spconv2.0 is not installed, the code will use mmcv's spconv op.

BC-breaking (Optional)

.

Checklist

• fix get_start docs
• fix ut
• add spconv check in collect_env.py

[codecov]

Codecov Report

Merging #1421 (ea9ab7a) into dev (276b264) will decrease coverage by 0.11%.
The diff coverage is 19.41%.

@@            Coverage Diff             @@
##              dev    #1421      +/-   ##
==========================================
- Coverage   50.86%   50.74%   -0.12%     
==========================================
  Files         217      219       +2     
  Lines       18567    18655      +88     
  Branches     3031     3053      +22     
==========================================
+ Hits         9444     9467      +23     
- Misses       8555     8617      +62     
- Partials      568      571       +3     

Flag	Coverage Δ
unittests	50.74% <19.41%> (-0.12%)	⬇️

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

Impacted Files	Coverage Δ
...det3d/ops/spconv/overwrite_spconv/write_spconv2.py	9.52% <9.52%> (ø)
mmdet3d/ops/sparse_block.py	46.96% <14.28%> (-3.04%)	⬇️
mmdet3d/models/middle_encoders/sparse_encoder.py	74.19% <50.00%> (-2.08%)	⬇️
mmdet3d/models/middle_encoders/sparse_unet.py	61.53% <50.00%> (-0.97%)	⬇️
mmdet3d/utils/collect_env.py	43.75% <50.00%> (+0.89%)	⬆️
mmdet3d/ops/spconv/__init__.py	66.66% <66.66%> (ø)
mmdet3d/datasets/pipelines/transforms_3d.py	88.25% <0.00%> (+0.85%)	⬆️

---

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 d842546...ea9ab7a. Read the comment docs.

[Tai-Wang]

Just need to fix minor comments.

[ZCMax]

seems we need to add unittest to ensure the coverage rate.

[ammaryasirnaich]

@VVsssssk, I have one question after training with the spconv2. If we use the mmdet3d/apis/inference.py for doing inference_detector(model, pcd) than which version will it choose spconV2 or V1? However, I think it's using V1 which is causing a dimensional mismatch with the trained weights!
I will really appreciate if you can confirm this, thanks