Add support for multiple random flips

[SBCV]

Motivation

This PR addresses #1781 by adding support for multiple RandomFlip operations in the training_pipeline.

Modification

With this PR it is possible to define multiple random flip operations by adding something like dict(type='RandomFlip', prob=(0.5, 0.5), direction=("vertical", "horizontal")) to the training_pipeline, e.g.

train_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='LoadAnnotations', reduce_zero_label=True),
    ...
    dict(type='RandomFlip', prob=(0.5, 0.5), direction=("vertical", "horizontal"))`
    ...
    dict(type='DefaultFormatBundle'),
    dict(type='Collect', keys=['img', 'gt_semantic_seg']),
]

Legacy inputs such as dict(type='RandomFlip', prob=0.5, direction="vertical" are still allowed (i.e. the patch is backward compatible).

Use cases (Optional)

For specific image domains it is reasonable to flip the data in both directions.

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues. [Yes]
2. The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness. [There are already unit tests covering the random flip operator. Is it necessary to adjust these?]
3. The documentation has been modified accordingly, like docstring or example tutorials. [Yes]

[codecov[bot]]

Codecov Report

Merging #1812 (1dd0bc1) into master (13d4c39) will increase coverage by 0.00%. The diff coverage is 95.65%.

@@           Coverage Diff           @@
##           master    #1812   +/-   ##
=======================================
  Coverage   89.04%   89.04%           
=======================================
  Files         144      144           
  Lines        8636     8648   +12     
  Branches     1458     1463    +5     
=======================================
+ Hits         7690     7701   +11     
  Misses        706      706           
- Partials      240      241    +1     

Flag	Coverage Δ
unittests	89.04% <95.65%> (+<0.01%)	:arrow_up:

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

Impacted Files	Coverage Δ
mmseg/datasets/pipelines/transforms.py	97.91% <95.65%> (-0.15%)	:arrow_down:

---

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 03c2447...1dd0bc1. Read the comment docs.

[SBCV]

See https://github.com/open-mmlab/mmsegmentation/pull/1918