[Enhance] Efficient implementation of PointSegClassMapping

[Wuziyi616]

The original implementation uses for-loop to convert every label, which is super slow in large scenes (e.g. S3DIS seg dataset without sub-sampling). I modify it to use matrix indexing, which could be faster. Matrix indexing is also a common practice in other 3D Seg repos.

[Wuziyi616]

I have carefully checked and modified all the places which use PointSegClassMapping in the codebase.

[codecov]

Codecov Report

Merging #489 (880ced4) into master (f7356f4) will increase coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #489      +/-   ##
==========================================
+ Coverage   50.62%   50.65%   +0.02%     
==========================================
  Files         191      191              
  Lines       14450    14452       +2     
  Branches     2367     2366       -1     
==========================================
+ Hits         7315     7320       +5     
+ Misses       6643     6640       -3     
  Partials      492      492              

Flag	Coverage Δ
unittests	50.65% <100.00%> (+0.02%)	⬆️

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

Impacted Files	Coverage Δ
mmdet3d/datasets/s3dis_dataset.py	90.69% <ø> (ø)
mmdet3d/datasets/scannet_dataset.py	94.56% <ø> (ø)
mmdet3d/datasets/pipelines/loading.py	85.03% <100.00%> (+1.21%)	⬆️
mmdet3d/models/dense_heads/fcos_mono3d_head.py	13.10% <0.00%> (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 f7356f4...880ced4. Read the comment docs.