TensorFlow, CUDA, cuDNN, OpenCV 4.5.5, Ubuntu 18.04

Dockerfile

ARG CUDA_VERSION=11.2.2
ARG CUDNN_VERSION=8
ARG UBUNTU_VERSION=18.04

# https://hub.docker.com/r/nvidia/cuda
FROM nvidia/cuda:${CUDA_VERSION}-cudnn${CUDNN_VERSION}-devel-ubuntu${UBUNTU_VERSION}

ARG PYTHON_VERSION=3.8
ARG OPENCV_VERSION=4.5.5

ENV DEBIAN_FRONTEND=noninteractive

# https://github.com/NVIDIA/nvidia-docker/issues/1632
# RUN rm /etc/apt/sources.list.d/cuda.list
# RUN rm /etc/apt/sources.list.d/nvidia-ml.list

ARG CUDA_SHORT_VERSION=11.2
# Compiling for NVIDIA A100
# https://study.marearts.com/2021/06/cudaarchbin-table-for-gpu-type.html
# https://en.wikipedia.org/wiki/CUDA
# https://developer.nvidia.com/cuda-gpus
ARG CUDA_ARCH=8.0
# https://www.tensorflow.org/install/source?hl=en#gpu
ARG TF_VERSION=2.7.0

ENV PATH=${PATH}:/usr/local/cuda-${CUDA_SHORT_VERSION}/bin
ENV LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/cuda-${CUDA_SHORT_VERSION}/lib64:/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64
ENV LIBRARY_PATH=${LIBRARY_PATH}:/usr/local/cuda-${CUDA_SHORT_VERSION}/lib64
ENV CPATH=${CPATH}:/usr/local/cuda/include

RUN apt-get -qq update && \
    apt-get -qq install  \
#   python :
        python${PYTHON_VERSION} \
        python${PYTHON_VERSION}-dev \
        libpython${PYTHON_VERSION} \
        libpython${PYTHON_VERSION}-dev \
        python-dev \
        python3-pip \
        python3-dev \
        python3-setuptools \
#   developement tools, opencv image/video/GUI dependencies, optimiztion packages , etc ...  :
        apt-utils \
        autoconf \
        automake \
        build-essential \
        checkinstall \
        cmake \
        g++ \
        gcc \
        gfortran \
        git \
        libatlas-base-dev \
        libavcodec-dev \
        libavformat-dev \
        libavresample-dev \
        libeigen3-dev \
        libexpat1-dev \
        libglew-dev \
        libgtk-3-dev \
        libjpeg-dev \
        libopenexr-dev \
        libpng-dev \
        libpostproc-dev \
        libpq-dev \
        libprotobuf-dev \
        libqt5opengl5-dev \
        libsm6 \
        libswscale-dev \
        libtbb2 \
        libtbb-dev \
        libtiff-dev \
        libtool \
        libv4l-dev \
        libwebp-dev \
        libxext6 \
        libxrender1 \
        libxvidcore-dev \
        ninja-build \
        pkg-config \
        protobuf-compiler \
        qt5-default \
        unzip \
        wget \
        yasm \
        zlib1g-dev \
#   GStreamer :
        libgstreamer1.0-0 \
        gstreamer1.0-plugins-base \
        gstreamer1.0-plugins-good \
        gstreamer1.0-plugins-bad \
        gstreamer1.0-plugins-ugly \
        gstreamer1.0-libav \
        gstreamer1.0-doc \
        gstreamer1.0-tools \
        gstreamer1.0-x \
        gstreamer1.0-alsa \
        gstreamer1.0-gl \
        gstreamer1.0-gtk3 \
        gstreamer1.0-qt5 \
        gstreamer1.0-pulseaudio \
        libgstreamer1.0-dev \
        libgstreamer-plugins-base1.0-dev && \
    rm -rf /var/lib/apt/lists/* && \
    apt-get purge   --auto-remove && \
    apt-get clean

# install new pyhton system wide :
RUN update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.6 1 && \
    update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 2 && \
    update-alternatives --config python3

# numpy for the newly installed python :
RUN wget https://bootstrap.pypa.io/get-pip.py  && \
    python${PYTHON_VERSION} get-pip.py --no-setuptools --no-wheel && \
    rm get-pip.py && \
    python${PYTHON_VERSION} -m pip uninstall -y numpy && \
    python${PYTHON_VERSION} -m pip install numpy==1.19.5

# opencv and opencv-contrib :
RUN cd /opt/ && \
    wget https://github.com/opencv/opencv/archive/${OPENCV_VERSION}.zip -O opencv.zip && \
    unzip -qq opencv.zip && \
    rm opencv.zip && \
    wget https://github.com/opencv/opencv_contrib/archive/${OPENCV_VERSION}.zip -O opencv-co.zip && \
    unzip -qq opencv-co.zip && \
    rm opencv-co.zip && \
    mkdir /opt/opencv-${OPENCV_VERSION}/build && cd /opt/opencv-${OPENCV_VERSION}/build && \
    # cmake \
    cmake -GNinja \
        -D BUILD_TESTS=OFF \
        -D BUILD_PERF_TESTS=OFF \
        -D BUILD_EXAMPLES=OFF \
        -D BUILD_opencv_java=OFF \
        -D CMAKE_LIBRARY_PATH=/usr/local/cuda/lib64/stubs \
        -D CMAKE_BUILD_TYPE=RELEASE \
        -D CUDA_FAST_MATH=ON \
        -D CUDA_ARCH_BIN=${CUDA_ARCH} \
        -D CUDA_ARCH_PTX=${CUDA_ARCH} \
        -D CUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda \
        -D ENABLE_FAST_MATH=ON \
        -D HAVE_opencv_python3=ON \
        -D INSTALL_C_EXAMPLES=OFF \
        -D OPENCV_DNN_CUDA=ON \
        -D OPENCV_ENABLE_NONFREE=ON \
        -D OPENCV_GENERATE_PKGCONFIG=ON \
        -D OPENCV_EXTRA_MODULES_PATH=/opt/opencv_contrib-${OPENCV_VERSION}/modules \
        -D OPENCV_PYTHON3_INSTALL_PATH=/usr/lib/python3/dist-packages \
        -D PYTHON_EXECUTABLE=/usr/bin/python3 \
        -D WITH_CUBLAS=ON \
        -D WITH_CUDA=ON \
        -D WITH_CUDNN=ON \
        -D WITH_CUFFT=ON \
        -D WITH_EIGEN=ON \
        -D WITH_IPP=ON \
        -D WITH_NVCUVID=ON \
        # -D WITH_OPENGL=ON \
        # -D WITH_QT=ON \
        # -D WITH_TBB=ON \
        # -D PYTHON2_EXECUTABLE=$(python${PYTHON_VERSION} -c "import sys; print(sys.prefix)") \
        # -D CMAKE_INSTALL_PREFIX=$(python${PYTHON_VERSION} -c "import sys; print(sys.prefix)") \
        # -D PYTHON_INCLUDE_DIR=$(python${PYTHON_VERSION} -c "from distutils.sysconfig import get_python_inc; print(get_python_inc())") \
        # -D PYTHON_PACKAGES_PATH=$(python${PYTHON_VERSION} -c "from distutils.sysconfig import get_python_lib; print(get_python_lib())") \
        .. &&\
    # make -j$(nproc) && \
    # make install && \
    ninja && \
    ninja install && \
    ldconfig && \
    rm -rf /opt/opencv-${OPENCV_VERSION} && rm -rf /opt/opencv_contrib-${OPENCV_VERSION}

ENV NVIDIA_DRIVER_CAPABILITIES all
ENV XDG_RUNTIME_DIR "/tmp"

RUN pip3 install testresources setuptools scikit-build

# numpy==1.19.5 is already installed before opencv compilation
RUN pip3 install -U cython pkgconfig wrapt==1.12 future==0.18.2 mock==3.0.5 keras_preprocessing==1.1.2 keras_applications==1.0.8 gast==0.4.0 protobuf==3.19.6 pybind11 pkgconfig packaging h5py==3.1.0 shapely easydict scipy requests gdown

RUN pip3 install gunicorn==20.1.0 celery==5.0.3 flask==2.0.3 redis==4.3.5 tensorflow==${TF_VERSION}

WORKDIR /app

RUN echo "alias python=python3" >> ~/.bashrc
RUN echo "alias pip=pip3" >> ~/.bashrc

COPY . ./

# Download files from GDrive
RUN gdown https://drive.google.com/uc?id=1zGnGkTvFAsUDacYaVcay3CFiZDkNXT0P
RUN mv yolo.zip ./yolo
RUN unzip -o ./yolo/yolo.zip -d ./yolo
RUN rm ./yolo/yolo.zip

# RUN gdown https://drive.google.com/uc?id=1HNg9cdWAXdJoc9aeBQb0-gaw0ZXOc9Rk
# RUN mv yolov4.onnx ./yolo

RUN gdown https://drive.google.com/uc?id=12afPUXgyJC0n-GhiSGRA5MOoC3GHC93x
RUN mv weights.pb.zip ./deep_sort
RUN unzip -o ./deep_sort/weights.pb.zip -d ./deep_sort
RUN rm ./deep_sort/weights.pb.zip

# ENTRYPOINT ["/bin/bash"]

CMD ["sh", "-c", "gunicorn -b 0.0.0.0:8000 wsgi:app --daemon && celery -A server.celery worker -E"]

Revision 2 works but it consumes 8GB of GPU memory for a single stream.