[Feature] Add S3DIS dataset for 3D object detection

[filaPro]

Close #817.

@Wuziyi616 can you have a look about general structure? Then I can also add unit test and documentation. I'm also attaching visualization from browse_dataset.

[codecov]

Codecov Report

Merging #835 (552773b) into master (fc9e0d9) will increase coverage by 0.05%.
The diff coverage is 75.00%.

@@            Coverage Diff             @@
##           master     #835      +/-   ##
==========================================
+ Coverage   49.18%   49.23%   +0.05%     
==========================================
  Files         210      210              
  Lines       15986    16019      +33     
  Branches     2552     2556       +4     
==========================================
+ Hits         7862     7887      +25     
- Misses       7625     7630       +5     
- Partials      499      502       +3     

Flag	Coverage Δ
unittests	49.23% <75.00%> (+0.05%)	⬆️

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

Impacted Files	Coverage Δ
mmdet3d/datasets/s3dis_dataset.py	88.57% <74.19%> (-6.10%)	⬇️
mmdet3d/datasets/__init__.py	100.00% <100.00%> (ø)
mmdet3d/datasets/pipelines/formating.py	60.00% <0.00%> (ø)
mmdet3d/models/detectors/single_stage_mono3d.py	16.66% <0.00%> (ø)
mmdet3d/datasets/pipelines/transforms_3d.py	90.46% <0.00%> (+0.04%)	⬆️

---

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 fc9e0d9...552773b. Read the comment docs.

[Wuziyi616]

Good job! @filaPro two questions:

• Similar to ScanNet detection, S3DIS detection also doesn't involve yaw prediction right? So the boxes here are all axis-aligned.
• Maybe we can move the two S3DIS detection datasets code to be above the two S3DIS segmentation datasets code? This follows scannet_dataset.py

The browse_dataset result is great! Only need to add some docstring and minor code structure refinement. Also unit tests are needed.

[Wuziyi616]

I let @THU17cyz have a look here because this may involve our progress of refactoring coordinate system.

[filaPro]

@Wuziyi616,

• Yes this dataset doesn't have yaw angle.
• Moved detection dataset to be above the segmentation one.

I also removed _S3DISDataset in favour of ConcatDataset. This also requires a fmall fix in browse_dataset. Do you think it is ok now?

[yezhen17]

Great work, many thanks to @filaPro ! Similar to ScanNet det, there is no conflict with our ongoing coordinate system refactor.

[Wuziyi616]

Almost ready to merge. Only need to add some unit tests.

[filaPro]

Don't quite understand lint problems, so revert yapf: disable back.

[Wuziyi616]

Don't quite understand lint problems, so revert yapf: disable back.

This is because sometimes isort and yapf will have different behaviors in formatting some codes. You need to disable one of them to pass the linting check.

[yezhen17]

Ready to merge after unit tests are added

[filaPro]

Also updated test data file for S3DIS.

[Wuziyi616]

Good Job @filaPro! Cheers!