A bug about dockerfile

Instructions To Reproduce the 🐛 Bug:

1. Full runnable code or full changes you made:

docker build -t detectron2:v0 .

1. What exact command you run: docker build -t detectron2:v0 .
2. Full logs or other relevant observations:

ERROR: Package 'detectron2' requires a different Python: 3.6.9 not in '>=3.7'

1. please simplify the steps as much as possible so they do not require additional resources to run, such as a private dataset.

Expected behavior:

it will build successful

Environment:

Provide your environment information using the following command:

Failed to load OpenCL runtime
No CUDA runtime is found, using CUDA_HOME='/usr/local/cuda'
---------------------  --------------------------------------------------------------------------
sys.platform           linux
Python                 3.6.9 (default, Mar 15 2022, 13:55:28) [GCC 8.4.0]
numpy                  1.13.3
detectron2             0.6 @/home/appuser/detectron2_repo/detectron2
detectron2._C          not built correctly: No module named 'detectron2._C'
Compiler ($CXX)        c++ (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0
DETECTRON2_ENV_MODULE  <not set>
PyTorch                1.10.0+cu111 @/home/appuser/.local/lib/python3.6/site-packages/torch
PyTorch debug build    False
GPU available          No: torch.cuda.is_available() == False
Pillow                 8.4.0
torchvision            0.11.1+cu111 @/home/appuser/.local/lib/python3.6/site-packages/torchvision
fvcore                 0.1.5
iopath                 0.1.10
cv2                    3.2.0
---------------------  --------------------------------------------------------------------------
PyTorch built with:
  - GCC 7.3
  - C++ Version: 201402
  - Intel(R) Math Kernel Library Version 2020.0.0 Product Build 20191122 for Intel(R) 64 architecture applications
  - Intel(R) MKL-DNN v2.2.3 (Git Hash 7336ca9f055cf1bfa13efb658fe15dc9b41f0740)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - LAPACK is enabled (usually provided by MKL)
  - NNPACK is enabled
  - CPU capability usage: AVX512
  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.1, CUDNN_VERSION=8.0.5, CXX_COMPILER=/opt/rh/devtoolset-7/root/usr/bin/c++, CXX_FLAGS= -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-sign-compare -Wno-unused-parameter -Wno-unused-variable -Wno-unused-function -Wno-unused-result -Wno-unused-local-typedefs -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.10.0, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON,

If your issue looks like an installation issue / environment issue, please first try to solve it yourself with the instructions in https://detectron2.readthedocs.io/tutorials/install.html#common-installation-issues

[farukcankaya]

As @peiyingxin said, root cause is python version inconsistency. I’ve tried two approaches to fix the problem:

• Upgrade ubuntu version from 18.04 to 20.4

  FROM nvidia/cuda:11.1.1-cudnn8-devel-ubuntu20.04 # 18.04 to 20.04
  ...
  RUN wget https://bootstrap.pypa.io/get-pip.py && \ # do not specify pip version
  ...
  

  It worked for building and running the image. issue#3524 says upgrading ubuntu causes opencv incompatibility problem but it might be outdated anymore.

• Upgrade python version on ubuntu 18.04 to Python3.7

   ...
   ENV DEBIAN_FRONTEND noninteractive
   RUN apt-get update && apt-get install -y \
        python3.7 python3.7-dev python3.7-distutils \
        python3-opencv ca-certificates git wget sudo ninja-build
  
    # Make python3 available for python3.7
    RUN update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.6 1
    RUN update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.7 2
    RUN update-alternatives --config python3
    # Make python available for python3.7
    RUN ln -sv /usr/bin/python3.7 /usr/bin/python
  
    # create a non-root user
    ...
    RUN wget https://bootstrap.pypa.io/pip/get-pip.py && \
        python3.7 get-pip.py --user && \
        rm get-pip.py
  
    # Important! Otherwise, it uses existing numpy from host-modules which throws error
    RUN pip install --user numpy==1.20.3
    ...
  

  This was a bit tricky. Even if the python version is upgraded to 3.7, detectron repo installation fetches the existing numpy==1.13.1 from host modules(/usr/lib/python3/dist-packages/) then it throws ImportError: cannot import name 'multiarray' from 'numpy.core' error. detectron2_numpy_import_error_full_log.txt

  To fix it, I simply added RUN pip install --user numpy==1.20.3. This installs numpy to local modules(./.local/lib/python3.7/site-packages) thus pip can fetch the correct numpy while installing detectron2 repo. detectron2_py37_successful_build_full_log.txt

  Finally, here is the full Dockerfile that worked for me: Dockerfile.txt

[Kodemannen]

The first of @farukcankaya’s bullets worked for me, but I had to also change PyTorch to be installed with the same version of Cuda as used in the Docker container.

I similarly changed the first line to FROM nvidia/cuda:11.1.1-cudnn8-devel-ubuntu20.04 , (installed the latest pip using RUN wget https://bootstrap.pypa.io/get-pip.py && \), then changed to /cu111/ (had /cu101/) in the line RUN pip install --user torch==1.10.1 torchvision==0.11.2 -f https://download.pytorch.org/whl/cu111/torch_stable.html