[Feature] Add Mosaic transform

[lkm2835]

Modification

mmseg/datasets/pipelines/transforms.py: Add class Mosaic
tests/test_data/test_transform.py: Add test_mosaic
mmseg/datasets/pipelines/__init__.py: import Mosaic

---

original code: Mosaic, test_mosaic in mmdet

[Junjun2016]

Hi @lkm2835
Thanks for your contribution again.
@RockeyCoss please review it and do some ablation studies.

[Junjun2016]

We may also need MultiImageMixDataset.

[Junjun2016]

We may also need MultiImageMixDataset.

Can adjust MultiImageMixDataset in mmdet (https://github.com/open-mmlab/mmdetection/blob/f08548bfd6d394a82566022709b5ce9e6b0a855e/mmdet/datasets/dataset_wrappers.py#L289) to mmseg accordingly.

[lkm2835]

We may also need MultiImageMixDataset.

I'll prepare and create another PR.

[Junjun2016]

We may also need MultiImageMixDataset.

I'll prepare and create another PR.

Great, it is better to add some results on your own datasets or mmseg supported datasets.

[codecov]

Codecov Report

Attention: Patch coverage is 97.72727% with 2 lines in your changes missing coverage. Please review.

Project coverage is 90.17%. Comparing base (7f06e01) to head (5bd9cb8).
Report is 254 commits behind head on master.

Files	Patch %	Lines
mmseg/datasets/pipelines/transforms.py	97.72%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1093      +/-   ##
==========================================
+ Coverage   90.07%   90.17%   +0.09%     
==========================================
  Files         125      125              
  Lines        7268     7356      +88     
  Branches     1206     1218      +12     
==========================================
+ Hits         6547     6633      +86     
- Misses        518      520       +2     
  Partials      203      203              

Flag	Coverage Δ
unittests	90.17% <97.72%> (+0.09%)	⬆️

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

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[RockeyCoss]

@lkm2835 Hi，The results are as follows:

exp_num	method	mIoU best	best index	mIoU last	last index
1	deeplabv3_r50-d8_512x512_160k_ade20k	0.43	10	0.43	10
2	deeplabv3_r50-d8_512x512_160k_ade20k_mosaic_0.2-1.8	0.4316	10	0.4316	10
3	deeplabv3_r50-d8_512x512_160k_ade20k_mosaic_0.5-1.5	0.4326	10	0.4326	10
4	deeplabv3_r50-d8_512x512_160k_ade20k_mosaic_0.8-1.2	0.4283	10	0.4283	10

We can see that

[lkm2835]

We have done different img scales test for Mosaic augmentation.

The results are as follows:
[Feature] Add MultiImageMixDataset #1105

[CLAassistant]

All committers have signed the CLA.