if cuda10 , I get an error at the end of the process. error: /home/dell/.cache/Python-Eggs/correlation_package-0.1-py2.7-linux-x86_64.egg-tmp/correlation_package/_ext/corr/_corr.so: undefined symbol: __cudaRegisterFatBinaryEnd /Anyone know how to solve this?

[wyhcqq]

No description provided.

[jeffbaena]

I am also forced to use CUDA 10 and I gave up for another implementation (only tried for inference) https://github.com/sniklaus/pytorch-pwc.
I would be happy if someone could help

[baowenbo]

I also got the same problem.
I can successfully compile the customized layers for pytorch with CUDA 10.1 on Ubuntu 18.04.
But I failed to import the _my_lib.so with undefined symbol: __cudaRegisterFatBinaryEnd error.

[jrenzhile]

In the readme file: https://github.com/NVlabs/PWC-Net/tree/master/PyTorch#installation we mentioned that the current implementation only work under CUDA 8.0

[rmbashirov]

Also had trouble with this issue.
I have successfully run script_pwc.py using this Dockerfile:

FROM nvidia/cuda:8.0-cudnn7-devel-ubuntu16.04

RUN apt-get update && apt-get install -y \
    rsync \
    htop \
    git \
    openssh-server \
    nano \
    cmake \
    python-opencv \
    python-pip
RUN pip install --upgrade pip

RUN apt-get -qq -y install curl bzip2 && \
    curl -sSL https://repo.continuum.io/miniconda/Miniconda2-4.6.14-Linux-x86_64.sh -o /tmp/miniconda.sh && \
    bash /tmp/miniconda.sh -bfp /usr/local && \
    rm -rf /tmp/miniconda.sh && \
    conda install -y python=2 && \
    conda update conda && \
    apt-get -qq -y autoremove && \
    apt-get autoclean && \
    rm -rf /var/lib/apt/lists/* /var/log/dpkg.log && \
    conda clean --all --yes

RUN conda install pytorch=0.2.0 cuda80 -c soumith
RUN pip install torchvision==0.2.2 visdom==0.1.8.8 dominate==2.3.5 opencv-python==4.1.0.25 cffi==1.12.2

[ZhengyiLuo]

Also had trouble with this issue.
I have successfully run script_pwc.py using this Dockerfile:

FROM nvidia/cuda:8.0-cudnn7-devel-ubuntu16.04

RUN apt-get update && apt-get install -y \
    rsync \
    htop \
    git \
    openssh-server \
    nano \
    cmake \
    python-opencv \
    python-pip
RUN pip install --upgrade pip

RUN apt-get -qq -y install curl bzip2 && \
    curl -sSL https://repo.continuum.io/miniconda/Miniconda2-4.6.14-Linux-x86_64.sh -o /tmp/miniconda.sh && \
    bash /tmp/miniconda.sh -bfp /usr/local && \
    rm -rf /tmp/miniconda.sh && \
    conda install -y python=2 && \
    conda update conda && \
    apt-get -qq -y autoremove && \
    apt-get autoclean && \
    rm -rf /var/lib/apt/lists/* /var/log/dpkg.log && \
    conda clean --all --yes

RUN conda install pytorch=0.2.0 cuda80 -c soumith
RUN pip install torchvision==0.2.2 visdom==0.1.8.8 dominate==2.3.5 opencv-python==4.1.0.25 cffi==1.12.2

Thanks for the Dockerfile; Still didn't work though. Thrown the same error on my end "undefined symbol: __cudaRegisterFatBinaryEnd". Using cuda 8.0 and everything.

[ZhengyiLuo]

Also had trouble with this issue.
I have successfully run script_pwc.py using this Dockerfile:

FROM nvidia/cuda:8.0-cudnn7-devel-ubuntu16.04

RUN apt-get update && apt-get install -y \
    rsync \
    htop \
    git \
    openssh-server \
    nano \
    cmake \
    python-opencv \
    python-pip
RUN pip install --upgrade pip

RUN apt-get -qq -y install curl bzip2 && \
    curl -sSL https://repo.continuum.io/miniconda/Miniconda2-4.6.14-Linux-x86_64.sh -o /tmp/miniconda.sh && \
    bash /tmp/miniconda.sh -bfp /usr/local && \
    rm -rf /tmp/miniconda.sh && \
    conda install -y python=2 && \
    conda update conda && \
    apt-get -qq -y autoremove && \
    apt-get autoclean && \
    rm -rf /var/lib/apt/lists/* /var/log/dpkg.log && \
    conda clean --all --yes

RUN conda install pytorch=0.2.0 cuda80 -c soumith
RUN pip install torchvision==0.2.2 visdom==0.1.8.8 dominate==2.3.5 opencv-python==4.1.0.25 cffi==1.12.2

Thanks for the Dockerfile; Still didn't work though. Thrown the same error on my end "undefined symbol: __cudaRegisterFatBinaryEnd". Using cuda 8.0 and everything.

I took that back. The issue I think, partially, is caused by the incompatible architecture. I changed sm_52 to the compatible sm_61 in make_cuda.sh (I am using an 1080 Ti) and it worked. Thanks!

[Italy2006]

No description provided.

请问你解决了这个问题吗？