[Fix] Unify camera poses

[filaPro]

Replace calib['K'] and calib['Rt'] with img_meta['depth2img'] for SUN RGB-D to be like img_meta['lidar2img'] for outdoor datasets. More discussion can be found in #620#issuecomment-859664418.

This PR affects visualization on SUN RGB-D and ImVoteNet model. I tested these 2 things. Validation mAP for SUN RGB-D improved from 64.04 to 64.61.

[filaPro]

Sorry, will think about failed tests later.

[codecov]

Codecov Report

Merging #653 (514b74b) into master (d61476a) will decrease coverage by 0.01%.
The diff coverage is 76.00%.

@@            Coverage Diff             @@
##           master     #653      +/-   ##
==========================================
- Coverage   51.02%   51.01%   -0.02%     
==========================================
  Files         204      204              
  Lines       15395    15383      -12     
  Branches     2492     2488       -4     
==========================================
- Hits         7856     7848       -8     
+ Misses       7021     7017       -4     
  Partials      518      518              

Flag	Coverage Δ
unittests	51.01% <76.00%> (-0.02%)	⬇️

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

Impacted Files	Coverage Δ
mmdet3d/apis/inference.py	54.95% <0.00%> (+1.32%)	⬆️
mmdet3d/core/bbox/box_np_ops.py	26.81% <ø> (ø)
mmdet3d/core/bbox/structures/coord_3d_mode.py	61.87% <ø> (-1.26%)	⬇️
mmdet3d/datasets/pipelines/formating.py	56.00% <ø> (ø)
mmdet3d/models/detectors/imvotenet.py	28.65% <0.00%> (ø)
mmdet3d/core/bbox/structures/utils.py	97.14% <100.00%> (+0.08%)	⬆️
mmdet3d/core/visualizer/image_vis.py	71.83% <100.00%> (-2.25%)	⬇️
mmdet3d/datasets/sunrgbd_dataset.py	75.55% <100.00%> (+0.84%)	⬆️
mmdet3d/models/fusion_layers/vote_fusion.py	93.33% <100.00%> (ø)
mmdet3d/datasets/utils.py	66.66% <0.00%> (-2.39%)	⬇️

---

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 d61476a...514b74b. Read the comment docs.

[Wuziyi616]

Good job! Overall the code is much cleaner and easy to follow now. Can you kindly post a few visualization of ImVoteNet so that we can make sure the correctness of this PR? (Also some browse_dataset results on SUN RGB-D multi-modality task if you have time)

[filaPro]

Also the CI is broken. Probably not by my commit.

[Wuziyi616]

Also the CI is broken. Probably not by my commit.

Yes it's because of MMSeg version. Will raise a PR to fix it.

[filaPro]

Hi @Wuziyi616 ,
I checked browse_dataset as you requested and fixed calib there. Now ImVoteNet visualization and browse_dataset visualization are the same for master and current branch. The results are in https://disk.yandex.ru/d/d_22eIbR_2yJyA .

[Wuziyi616]

Hi @filaPro, please merge master so that the CI failure caused by MMSeg version incompatibility will be solved. Then we can run CI to check your PR's correctness. Thanks.

[filaPro]

@Wuziyi616 Done.
@THU17cyz Fixed your comments.

[Wuziyi616]

Great work!