ROI Pooling Layer

[varunagrawal]

Code for ROI Pooling layer.
(#477)

[fmassa]

Hi @varunagrawal ,

Thanks a lot for the PR!
Could you first send the PR to the layers branch? I'd rather not merge this into master yet, as it will add a few dependencies that I'd rather double check throughly.

[varunagrawal]

@fmassa I took out the NMS and ROI Align layers in order to minimize the dependencies on this PR and tackle each thing separately. I don't know if this branch would merge with your layers branch.
Any recommendations on what should I do?

[fmassa]

When you create a PR, you can select to which branch you want it to send the PR to, see https://help.github.com/articles/creating-a-pull-request/ for some more information.

[fmassa]

Just saw your other PR. I think that's the way to go, I'd look into updating the layers branch.
Also, one more thing, what version of PyTorch did you use to compile?
With PyTorch pulled from master yesterday, I'm facing issues such as

/private/home/fmassa/.conda/envs/detectron/lib/python3.6/site-packages/torch/lib/include/THC/THCNumerics.cuh(163): error: more than one operator "<" matches these operands:
            built-in operator "arithmetic < arithmetic"
            function "operator<(const __half &, const __half &)"
            operand types are: at::Half < at::Half

/private/home/fmassa/.conda/envs/detectron/lib/python3.6/site-packages/torch/lib/include/THC/THCNumerics.cuh(167): error: more than one operator "<=" matches these operands:
            built-in operator "arithmetic <= arithmetic"
            function "operator<=(const __half &, const __half &)"
            operand types are: at::Half <= at::Half

/private/home/fmassa/.conda/envs/detectron/lib/python3.6/site-packages/torch/lib/include/THC/THCNumerics.cuh(171): error: more than one operator ">" matches these operands:
            built-in operator "arithmetic > arithmetic"
            function "operator>(const __half &, const __half &)"
            operand types are: at::Half > at::Half

/private/home/fmassa/.conda/envs/detectron/lib/python3.6/site-packages/torch/lib/include/THC/THCNumerics.cuh(175): error: more than one operator ">=" matches these operands:
            built-in operator "arithmetic >= arithmetic"
            function "operator>=(const __half &, const __half &)"
            operand types are: at::Half >= at::Half

[varunagrawal]

@fmassa so do you want me to just base off the original layers branch (without the updates from master) and update ROI pooling accordingly? Just want to make sure I understand your last comment correctly.

Also, regarding that compilation error, it is a regression introduced by pytorch/pytorch#10301. Syed should have a fix in by next week.

[fmassa]

Oh yes, I had seen the diff but I didn't check the discussion in the PR.

I think it's easier to base the diff off from the layers branch, and then I separately rebase layers on top of master in a follow-up commit. what do you think?

[varunagrawal]

That should work. I'll force-update the PR to align with the layers branch.

[varunagrawal]

@fmassa it is done. Please refer to #592.