Does this work for 0.4.1 / how to install?

[andrewjong]

Just checking if this works with PT 0.4.1 as well, because the README only specifies 0.4. I ask because I noticed the PyTorch 0.4.1 release tag includes "Breaking changes". Thanks!

[andrewjong]

I get the following compile error when running ./travis.sh

.conda/envs/pt0.4/lib/python3.7/site-packages/torch/utils/ffi/../../lib/include/THC/THCGeneral.h:12:10: 
fatal error: cuda.h: No such file or directory
 #include "cuda.h"

[andrewjong]

Nevermind, got it to compile by symlinking the include files from my cuda install location.

ln -s /opt/cuda/include/* ~/.conda/envs/pt0.4/lib/python3.7/site-packages/torch/utils/ffi/../../lib/include/THC/

[andrewjong]

I now get this error in python after running travis.sh:

>>> from prroi_pool import PrRoIPool2D
Traceback (most recent call last):
  File "/home/user/Development/PreciseRoIPooling-master/pytorch/prroi_pool/functional.py", line 16, in <module>
    from . import _prroi_pooling
  File "/home/user/Development/PreciseRoIPooling-master/pytorch/prroi_pool/_prroi_pooling/__init__.py", line 3, in <module>
    from .__prroi_pooling import lib as _lib, ffi as _ffi
ImportError: /home/user/Development/PreciseRoIPooling-master/pytorch/prroi_pool/_prroi_pooling/__prroi_pooling.so: undefined symbol: __cudaPopCallConfiguration

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/user/Development/PreciseRoIPooling-master/pytorch/prroi_pool/__init__.py", line 12, in <module>
    from .prroi_pool import *
  File "/home/user/Development/PreciseRoIPooling-master/pytorch/prroi_pool/prroi_pool.py", line 14, in <module>
    from .functional import prroi_pool2d
  File "/home/user/Development/PreciseRoIPooling-master/pytorch/prroi_pool/functional.py", line 18, in <module>
    raise ImportError('Can not found the compiled Precise RoI Pooling library. Run ./travis.sh in the directory first.')
ImportError: Can not found the compiled Precise RoI Pooling library. Run ./travis.sh in the directory first.

[vacancy]

Hi Andrew. I think this is not a problem with pytorch but your cuda installation. Could you please check if the version of your cuda runtime libraries matches the compile-time headers/libraries?

[lianxintao]

I also encountered the same problem, I don't know if this is a common problem.

[lianxintao]

ImportError: /home/user/Development/PreciseRoIPooling-master/pytorch/prroi_pool/_prroi_pooling/__prroi_pooling.so: undefined symbol: __cudaPopCallConfiguration>
I don't know how to solve this problem.

[deepbrainme]

Hi,I can't compile this library , could you give me some advice about this?

Creating extension directory /tmp/torch_extensions/prroi_pooling...
Detected CUDA files, patching ldflags
Emitting ninja build file /tmp/torch_extensions/prroi_pooling/build.ninja...
Building extension module prroi_pooling...
[1/3] /usr/local/cuda/bin/nvcc -DTORCH_EXTENSION_NAME=prroi_pooling -DTORCH_API_INCLUDE_EXTENSION_H -isystem /home/ning/miniconda3/envs/ATOM/lib/python3.6/site-packages/torch/lib/include -isystem /home/ning/miniconda3/envs/ATOM/lib/python3.6/site-packages/torch/lib/include/torch/csrc/api/include -isystem /home/ning/miniconda3/envs/ATOM/lib/python3.6/site-packages/torch/lib/include/TH -isystem /home/ning/miniconda3/envs/ATOM/lib/python3.6/site-packages/torch/lib/include/THC -isystem /usr/local/cuda/include -isystem /home/ning/miniconda3/envs/ATOM/include/python3.6m -D_GLIBCXX_USE_CXX11_ABI=0 -D__CUDA_NO_HALF_OPERATORS -D__CUDA_NO_HALF_CONVERSIONS -D__CUDA_NO_HALF2_OPERATORS__ --compiler-options '-fPIC' -std=c++11 -c /home/ning/code/atom/pytracking/ltr/external/PreciseRoIPooling/pytorch/prroi_pool/src/prroi_pooling_gpu_impl.cu -o prroi_pooling_gpu_impl.cuda.o
[2/3] c++ -MMD -MF prroi_pooling_gpu.o.d -DTORCH_EXTENSION_NAME=_prroi_pooling -DTORCH_API_INCLUDE_EXTENSION_H -isystem /home/ning/miniconda3/envs/ATOM/lib/python3.6/site-packages/torch/lib/include -isystem /home/ning/miniconda3/envs/ATOM/lib/python3.6/site-packages/torch/lib/include/torch/csrc/api/include -isystem /home/ning/miniconda3/envs/ATOM/lib/python3.6/site-packages/torch/lib/include/TH -isystem /home/ning/miniconda3/envs/ATOM/lib/python3.6/site-packages/torch/lib/include/THC -isystem /usr/local/cuda/include -isystem /home/ning/miniconda3/envs/ATOM/include/python3.6m -D_GLIBCXX_USE_CXX11_ABI=0 -fPIC -std=c++11 -c /home/ning/code/atom/pytracking/ltr/external/PreciseRoIPooling/pytorch/prroi_pool/src/prroi_pooling_gpu.c -o prroi_pooling_gpu.o
[3/3] c++ prroi_pooling_gpu.o prroi_pooling_gpu_impl.cuda.o -shared -L/usr/local/cuda/lib64 -lcudart -o _prroi_pooling.so
Loading extension module _prroi_pooling...
Can not compile Precise RoI Pooling library.
Done

[vacancy]

Hi @DeepBrainsMe This looks like a cuda setup problem. Can you go to https://github.com/vacancy/PreciseRoIPooling/blob/master/pytorch/prroi_pool/functional.py#L24-L25 and see why an ImportError is raised?

[deepbrainme]

Thanks for your answer， I have compiled it just now ,however a new bug arise.The GPU Don't execute .My GPU is 2080ti。GPU driver is 418，cuda9。请问这个prpooling程序对显卡有特殊的要求么？我怀疑是不是我的显卡版本问题。另外使用的是虚拟环境下的pytorch 0.4.1。 发自我的iPhone

…

------------------ Original ------------------ From: Jiayuan Mao <notifications@github.com> Date: Fri,Apr 26,2019 3:18 AM To: vacancy/PreciseRoIPooling <PreciseRoIPooling@noreply.github.com> Cc: Jesse Wang <1098069058@qq.com>, Mention <mention@noreply.github.com> Subject: Re: [vacancy/PreciseRoIPooling] Does this work for 0.4.1 / how to install? (#12) Hi @DeepBrainsMe This looks like a cuda setup problem. Can you go to https://github.com/vacancy/PreciseRoIPooling/blob/master/pytorch/prroi_pool/functional.py#L24-L25 and see why an ImportError is raised? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

[vacancy]

Hi @DeepBrainsMe It seems that this is a bug with other functions you are using in your code. Perhaps this issue is related? facebookresearch/Detectron#260

Also, can you run a small script to test whether the PrRoIPooling are working? Here is a small snippet:

        pool = PrRoIPool2D(2, 2, spatial_scale=0.5)

        features = torch.rand((4, 2, 24, 32)).cuda()
        rois = torch.tensor([
            [0, 0, 0, 4, 4],
            [1, 14, 14, 18, 18],
        ]).float().cuda()
        features.requires_grad = rois.requires_grad = True
        out = pool(features, rois)

        loss = out.sum()
        loss.backward()

[Hevans123]

Hi, @DeepBrainsMe @vacancy I meet some problem when I compile this library.It seems that there is something wrong using Ninja-build. Could you give me some advice. Thanks.
[1/3] /usr/local/cuda/bin/nvcc -DTORCH_EXTENSION_NAME=prroi_pooling -DTORCH_API_INCLUDE_EXTENSION_H -isystem ....... -D_GLIBCXX_USE_CXX11_ABI=0 -D__CUDA_NO_HALF_OPERATORS_ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --compiler-options '-fPIC' -std=c++11 -c /../PreciseRoIPooling/pytorch/prroi_pool/src/prroi_pooling_gpu_impl.cu -o prroi_pooling_gpu_impl.cuda.o
FAILED: prroi_pooling_gpu_impl.cuda.o
error: expected a declaration
1 error detected in the compilation of "/tmp/tmpxft_0000becb_00000000-6_prroi_pooling_gpu_impl.cpp1.ii".
......
......
ninja: build stopped: subcommand failed.
Error building extension '_prroi_pooling'
Done