Dockerfile.opencv

# syntax=docker/dockerfile:1

# Comments are provided throughout this file to help you get started.
# If you need more help, visit the Dockerfile reference guide at
# https://docs.docker.com/go/dockerfile-reference/

#! IMPORTANT: Please make sure that your NVIDIA Jetson device is running the latest JetPack version.
#! This is important since the image mounts various libraries (e.g., libcudnn, CUDA) from the host system.

#! NOTE: This image builds OpenCV without the NVIDIA Video Codec SDK by default due to copyright issues.
#! Thus, the image provided at ghcr.io/appsolves/lanepilot/opencv_base:latest does not include the necessary dependencies.
#! To build the image with the NVIDIA Video Codec SDK, do the following:
#! 1. Create a directory named video_codec_sdk in the same directory as this Dockerfile.
#! 2. Download the NVIDIA Video Codec SDK zip file and place it in the video_codec_sdk directory.
#! 3. Run the `scripts/build_opencv.sh` script to build the image locally with the NVIDIA Video Codec SDK included.

ARG L4T_TRT_VERSION=r10.3.0
ARG BASE_IMAGE_VERSION=r36.2.0
ARG PLATFORM=linux/arm64

### Builder image ###
FROM --platform=${PLATFORM} nvcr.io/nvidia/l4t-tensorrt:${L4T_TRT_VERSION}-devel AS opencv-builder

# Compute capability is set for NVIDIA Jetson Orin Nano by default
ARG CUDA_COMPUTE_CAPABILITY=8.7
# Install OpenCV Contrib modules
ARG OPENCV_CONTRIB=1
# Set to non-interactive mode to avoid prompts during package installation
ENV DEBIAN_FRONTEND=noninteractive

# Install dependencies and required packages
RUN apt-get update -y --fix-missing && apt-get upgrade -y --fix-missing
RUN apt-get install -y build-essential ninja-build git wget unzip pkg-config zlib1g-dev \
    python3-dev python-dev-is-python3 \
    gstreamer1.0-tools libgstreamer-plugins-base1.0-dev \
    libgstreamer-plugins-good1.0-dev \
    libtbb2 cmake libjpeg8-dev libjpeg-turbo8-dev \
    libglew-dev python3-pip libtiff-dev \
    libavcodec-dev libavformat-dev libswscale-dev libswresample-dev \
    libxvidcore-dev libx264-dev \
    libtbb-dev libxine2-dev \
    libv4l-dev v4l-utils qv4l2 \
    libtesseract-dev libpostproc-dev \
    libvorbis-dev libaom-dev libdav1d-dev \
    libyuv-dev libjpeg-dev libpng-dev \
    libfaac-dev libmp3lame-dev libtheora-dev \
    libopencore-amrnb-dev libopencore-amrwb-dev \
    libopenblas-dev libatlas-base-dev libblas-dev \
    liblapack-dev liblapacke-dev libeigen3-dev gfortran \
    libhdf5-dev libprotobuf-dev protobuf-compiler \
    libgoogle-glog-dev libgflags-dev && \
    apt-get clean && rm -rf /var/lib/apt/lists/*

# Install libavif
WORKDIR $HOME
RUN git clone https://github.com/AOMediaCodec/libavif.git \
    && cd libavif && mkdir build && cd build \
    && cmake .. -DCMAKE_BUILD_TYPE=Release -G Ninja \
    && ninja && ninja install \
    && cd $HOME \
    && rm -rf $HOME/libavif

# Install cuDNN
RUN wget https://developer.download.nvidia.com/compute/cudnn/redist/cudnn/linux-$(uname -m)/cudnn-linux-$(uname -m)-9.10.0.56_cuda12-archive.tar.xz \
    && tar -xvf cudnn-linux-$(uname -m)-9.10.0.56_cuda12-archive.tar.xz \
    && cd cudnn-linux-$(uname -m)-9.10.0.56_cuda12-archive \
    && cp include/cudnn*.h /usr/local/cuda/include \
    && cp lib/libcudnn* /usr/local/cuda/lib64 \
    && cd $HOME \
    && rm -rf $HOME/cudnn-linux-$(uname -m)-9.10.0.56_cuda12-archive \
    && rm -rf $HOME/cudnn-linux-$(uname -m)-9.10.0.56_cuda12-archive.tar.xz

# Clone OpenCV-Python
WORKDIR /opt
RUN git clone --recursive https://github.com/opencv/opencv-python.git
WORKDIR /opt/opencv-python
RUN git submodule update --init --recursive

# Fool git
RUN git config --global user.email "test@nonexistent.com" && \
    git config --global user.name "Test User"

# Update NvEncoder.cpp for OpenCV and commit changes
RUN wget "https://raw.githubusercontent.com/opencv/opencv_contrib/refs/heads/4.x/modules/cudacodec/src/NvEncoder.cpp" -O opencv_contrib/modules/cudacodec/src/NvEncoder.cpp
WORKDIR /opt/opencv-python/opencv_contrib
RUN git add modules/cudacodec/src/NvEncoder.cpp && git commit -m "Update cudacodec submodule to latest version"
WORKDIR /opt/opencv-python
RUN git add opencv_contrib && git commit -m "Updated submodule pointer to latest custom version"

# Make sure to build for numpy>=2.0.0
RUN pip install --no-cache-dir -U pip && \
    pip install --no-cache-dir -U "numpy>=2.0.0"

# Check NumPy version
RUN python3 -c "import numpy; print('NumPy for build:', numpy.__version__, '| Include:', numpy.get_include())"

# Set environment variables for OpenCV build
ENV ENABLE_CONTRIB=${OPENCV_CONTRIB}
ENV ENABLE_HEADLESS=1
ENV CFLAGS="--std=c++17"
ENV CXXFLAGS="--std=c++17"
ENV CMAKE_ARGS="-D CMAKE_BUILD_TYPE=Release \
    -D WITH_OPENCL=ON \
    -D CUDA_ARCH_BIN=${CUDA_COMPUTE_CAPABILITY} \
    -D CUDA_ARCH_PTX=${CUDA_COMPUTE_CAPABILITY} \
    -D WITH_CUDA=ON \
    -D WITH_CUDNN=ON \
    -D CUDNN_INCLUDE_DIR='/usr/local/cuda/include' \
    -D CUDNN_LIBRARY='/usr/local/cuda/lib64/libcudnn.so' \
    -D WITH_CUBLAS=ON \
    -D WITH_AVIF=ON \
    -D WITH_JPEG=ON \
    -D BUILD_JPEG=OFF \
    -D ENABLE_FAST_MATH=ON \
    -D CUDA_FAST_MATH=ON \
    -D CMAKE_CUDA_STANDARD=17 \
    -D CMAKE_CUDA_STANDARD_REQUIRED=ON \
    -D OPENCV_DNN_CUDA=ON \
    -D WITH_QT=OFF \
    -D WITH_VTK=OFF \
    -D WITH_OPENMP=ON \
    -D BUILD_TIFF=ON \
    -D WITH_FFMPEG=ON \
    -D WITH_TBB=ON \
    -D BUILD_TESTS=OFF \
    -D WITH_EIGEN=ON \
    -D WITH_PROTOBUF=ON \
    -D OPENCV_ENABLE_NONFREE=ON \
    -D INSTALL_C_EXAMPLES=OFF \
    -D INSTALL_PYTHON_EXAMPLES=OFF \
    -D OPENCV_GENERATE_PKGCONFIG=ON \
    -D BUILD_EXAMPLES=OFF \
    -D BUILD_opencv_java=OFF \
    -D CMAKE_CXX_STANDARD=17 \
    -D CMAKE_CXX_STANDARD_REQUIRED=ON \
    -D CMAKE_VERBOSE_MAKEFILE=ON \
    -D WITH_PYTHON=ON \
    -D WITH_NUMPY=ON \
    -D OPENCV_GAPI_GSTREAMER=ON \
    -D WITH_GSTREAMER=ON"

# Install NVIDIA Video Codec SDK
COPY . /tmp
RUN if [ -d /tmp/video_codec_sdk ] && ls /tmp/video_codec_sdk/*.zip 1>/dev/null 2>&1; then \
    mkdir -p /usr/local/cuda/include/ && \
    unzip /tmp/video_codec_sdk/*.zip -d /tmp/video_codec_sdk && \
    rm /tmp/video_codec_sdk/*.zip && \
    cp -rf /tmp/video_codec_sdk/*/Lib/linux/stubs/$(uname -m)/*.so /usr/local/cuda/lib64/ && \
    cp -rf /tmp/video_codec_sdk/*/Interface/*.h /usr/local/cuda/include/ && \
    echo "NVIDIA Video Codec SDK installed successfully." && \
    export CUDA_CODEC_ENABLED=1 ; \
    else \
    mkdir -p /tmp/video_codec_sdk && \
    echo "No video_codec_sdk zip found, skipping NVIDIA Video Codec SDK installation." && \
    export CUDA_CODEC_ENABLED=0 ; \
    fi

# Ensure that the NVIDIA Video Codec SDK is available
ENV LD_LIBRARY_PATH=/usr/local/cuda/lib64/:${LD_LIBRARY_PATH:-}
RUN LIBCUDA=$(find / -name libcuda.so -print -quit) && \
    if [ -f "$LIBCUDA" ]; then \
    ln -sf "$LIBCUDA" /usr/local/cuda/lib64/libcuda.so && ldconfig; \
    else \
    echo "ERROR: libcuda.so not found!" && exit 1; \
    fi

# Build OpenCV
RUN pip wheel . --verbose --wheel-dir /opt/opencv-install

### Final image ###
FROM --platform=${PLATFORM} nvcr.io/nvidia/l4t-base:${BASE_IMAGE_VERSION} AS base

LABEL org.opencontainers.image.authors="contact@appsolves.dev"
LABEL org.opencontainers.image.source=https://github.com/AppSolves/LanePilot
LABEL org.opencontainers.image.description="An image that contains and/or mounts CUDA, CUDNN, TensorRT, PyTorch, OpenCV (Python) and GStreamer runtime dependencies."
LABEL org.opencontainers.image.licenses=https://raw.githubusercontent.com/AppSolves/LanePilot/refs/heads/v2/LICENSE.md

# Set default python version to 3.10
ARG PYTHON_VERSION=3.10
ARG TORCH_INSTALL=https://developer.download.nvidia.cn/compute/redist/jp/v61/pytorch/torch-2.5.0a0+872d972e41.nv24.08.17622132-cp310-cp310-linux_aarch64.whl

# Update environment variables
ENV PATH=/usr/local/bin:${PATH:-}
ENV LD_LIBRARY_PATH=/hostlibs/all:/hostlibs/cuda/lib64/:/usr/local/cuda/lib64/:${LD_LIBRARY_PATH:-}
ENV PKG_CONFIG_PATH=/usr/local/lib/pkgconfig:${PKG_CONFIG_PATH:-}
ENV PYTHONPATH=/usr/local/python:${PYTHONPATH:-}

# Update dependencies
RUN ldconfig && \
    apt-get update -y --fix-missing && \
    apt-get upgrade -y --fix-missing && \
    apt-get install -y --no-install-recommends \
    gstreamer1.0-plugins-base \
    gstreamer1.0-plugins-good \
    gstreamer1.0-plugins-bad \
    gstreamer1.0-plugins-ugly \
    gstreamer1.0-libav \
    gstreamer1.0-tools \
    gstreamer1.0-x \
    gstreamer1.0-alsa \
    gstreamer1.0-gl \
    gstreamer1.0-pulseaudio \
    libhdf5-103 \
    libgoogle-glog0v5 \
    libtesseract4 \
    libtbb12 \
    libatlas3-base \
    libyuv0 \
    iproute2 \
    gosu \
    python3 python3-pip

# Copy NVIDIA Video Codec SDK
COPY --from=opencv-builder /tmp/video_codec_sdk /tmp/video_codec_sdk
RUN if [ -d /tmp/video_codec_sdk ] && ls /tmp/video_codec_sdk/*.so 1>/dev/null 2>&1; then \
    mkdir -p /usr/local/cuda/include/ && \
    mkdir -p /usr/local/cuda/lib64/ && \
    cp -rf /tmp/video_codec_sdk/*/Lib/linux/stubs/$(uname -m)/*.so /usr/local/cuda/lib64/ && \
    cp -rf /tmp/video_codec_sdk/*/Interface/*.h /usr/local/cuda/include/ && \
    echo "NVIDIA Video Codec SDK installed successfully." && \
    export CUDA_CODEC_ENABLED=1 ; \
    else \
    echo "No video_codec_sdk zip found, skipping NVIDIA Video Codec SDK installation." ; \
    export CUDA_CODEC_ENABLED=0 ; \
    fi

# Copy libavif
COPY --from=opencv-builder /usr/local/lib/libavif* /usr/local/lib/

# Copy OpenCV and python bindings
COPY --from=opencv-builder /opt/opencv-install /opt/opencv-install
WORKDIR /
RUN pip install --no-cache-dir -U pip && \
    opencv_wheel=$(ls /opt/opencv-install/opencv_contrib_python_headless-*.whl | head -n1) && \
    pip install --force-reinstall --upgrade --no-cache-dir "$opencv_wheel" && \
    rm -rf /opt/opencv-install

# Install PyTorch
RUN pip install --no-cache-dir -U ${TORCH_INSTALL}

# Copy root files
COPY --from=root /LICENSE.md /LICENSE.md

# Clean up
RUN apt-get clean && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*

# Remove headers, static libs, pkgconfig, docs and other dev stuff to save space
RUN rm -rf /usr/local/include \
    /usr/local/lib/*.a \
    /usr/local/lib/*.la \
    /usr/local/lib/*.cmake \
    /usr/local/lib/pkgconfig \
    /usr/local/lib/cmake \
    /usr/local/share \
    /usr/local/doc

# Strip binaries to reduce size
RUN find /usr/local -type f -executable ! -name '*.so*' -exec strip --strip-unneeded {} + 2>/dev/null || true

# Remove pip cache and __pycache__
RUN rm -rf /root/.cache /root/.pip /usr/local/lib/python*/dist-packages/*/__pycache__

# Set the default command to run when starting the container
CMD ["python3", "-c", "import cv2; print(cv2.getBuildInformation())"]