Add RROIAlign

[ElaineBao]

Description

Add RROIAlign operator and test case.
Different from ROI Align, RROI Align uses rotated rois, which is suitable for text detection.
Reference paper: Ma, Jianqi, et al. "Arbitrary-Oriented Scene Text Detection via Rotation Proposals."
IEEE Transactions on Multimedia, 2018.

Checklist

Essentials

Please feel free to remove inapplicable items for your PR.

• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage:
• Unit tests are added for small changes to verify correctness (e.g. adding a new operator)
• To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change

Changes

• RROIAlign
• test case for RROIAlign

Comments

• Only forward pass of RROIAlign is implemented.

[pengzhao-intel]

@wkcn could you help to take a review?
FYI, the backward will be submitted later.

[pengzhao-intel]

Could you provide some performance data?

[ElaineBao]

Could you provide some performance data?

Could you give a hint on what kind of performance data can be provided?

[pengzhao-intel]

Could you provide some performance data?

Could you give a hint on what kind of performance data can be provided?

A simple way is to compare the performance with and without your omp pragma and see how much speedup from your parallelization.

[pengzhao-intel]

btw, I think you still need to register CUDA version and error out with the hint like "No implementation"

[ElaineBao]

Could you provide some performance data?

Could you give a hint on what kind of performance data can be provided?

A simple way is to compare the performance with and without your omp pragma and see how much speedup from your parallelization.

Hi, @pengzhao-intel, the performance w/o omp is as follows:

omp option	performance
with omp	0.0586 s
without omp	0.4895 s
speedup	8.35X

Settings for the performance data:
clx-8260L, N=64, C=512, H=16, W=16, num_rois=1000, iterations=1000

[wkcn]

Great work! Thank you! LGTM : )

[pengzhao-intel]

@ciyongch could you take a look as well?

[pengzhao-intel]

Thanks for your contribution.