如何在l4t-base:r32.5.0 Docker镜像中从源码编译Open3D？

完整可运行Dockerfile #

FROM dustynv/ros:foxy-ros-base-l4t-r32.5.0
WORKDIR /workspace

# 安装基础编译依赖
RUN apt-get update && apt-get install -y --no-install-recommends \
    build-essential \
    gfortran \
    libopenblas-dev \
    liblapack-dev \
    liblapacke-dev \
    xorg-dev \
    libglu1-mesa-dev \
    wget \
    && rm -rf /var/lib/apt/lists/*

##########################
### CMAKE安装配置 ###
##########################
RUN wget https://github.com/Kitware/CMake/releases/download/v3.20.6/cmake-3.20.6-linux-aarch64.tar.gz && \
    tar -xzf cmake-3.20.6-linux-aarch64.tar.gz && \
    mv cmake-3.20.6-linux-aarch64 cmake && \
    rm cmake-3.20.6-linux-aarch64.tar.gz

# 全局暴露cmake命令，适配Docker非交互执行环境
ENV PATH="/workspace/cmake/bin:${PATH}"

#################################
### Open3D 0.13.0编译安装 ###
#################################
RUN wget https://github.com/isl-org/Open3D/archive/refs/tags/v0.13.0.tar.gz && \
    tar -xzf v0.13.0.tar.gz && \
    mv Open3D-0.13.0 open3d && \
    rm v0.13.0.tar.gz

RUN cd open3d && mkdir build && cd build && \
    cmake \
    -DCMAKE_BUILD_TYPE=Release \
    -DBUILD_SHARED_LIBS=ON \
    -DBUILD_CUDA_MODULE=OFF \
    -DBUILD_GUI=OFF \
    -DBUILD_TENSORFLOW_OPS=OFF \
    -DBUILD_PYTORCH_OPS=OFF \
    -DBUILD_UNIT_TESTS=OFF \
    -DCMAKE_INSTALL_PREFIX=/usr/local/open3d_install \
    -DPYTHON_EXECUTABLE=/usr/bin/python3 \
    # ARM架构专属编译参数，修复对应报错
    -DCMAKE_ASM_NASM_COMPILER=FALSE \
    -DCMAKE_C_FLAGS=-pthread \
    -DCMAKE_CXX_FLAGS=-pthread \
    -DCMAKE_EXE_LINKER_FLAGS="-lpthread -lm" \
    -DBUILD_SIMD_NEON=ON \
    -DBUILD_WEBRTC=OFF \
    -DBUILD_LIBREALSENSE=OFF \
    .. && \
    make -j$(nproc) && \
    make install

# 配置Open3D全局环境变量
ENV PATH="/usr/local/open3d_install/bin:${PATH}"
ENV LD_LIBRARY_PATH="/usr/local/open3d_install/lib:${LD_LIBRARY_PATH}"
ENV PYTHONPATH="/usr/local/open3d_install/lib/python3.8/site-packages:${PYTHONPATH}"

关键修改说明 #

• 补全基础编译依赖和线性代数库：新增安装build-essential、gfortran、libopenblas-dev、liblapack-dev、liblapacke-dev，解决BLAS、LAPACK相关函数未定义的报错
• 修复cmake命令生效问题：将原有的alias配置改为通过ENV全局暴露cmake路径，适配Docker非交互执行环境的特性
• 跳过x86专属NASM汇编检测：新增-DCMAKE_ASM_NASM_COMPILER=FALSE参数，ARM架构无需NASM编译器，直接跳过检测逻辑
• 显式指定pthread链接参数：新增-DCMAKE_C_FLAGS=-pthread、-DCMAKE_CXX_FLAGS=-pthread、-DCMAKE_EXE_LINKER_FLAGS="-lpthread -lm"参数，解决pthread函数未定义的链接错误
• 优化ARM架构适配：新增-DBUILD_SIMD_NEON=ON启用ARM专属NEON指令集优化，新增-DBUILD_WEBRTC=OFF、-DBUILD_LIBREALSENSE=OFF禁用兼容性差的非必要模块，降低编译复杂度

编译运行注意事项 #

• Jetson Xavier NX设备编译时建议提前开启至少4G交换分区，避免内存不足导致编译中断
• 若需要CUDA加速支持，可将-DBUILD_CUDA_MODULE=OFF参数修改为-DBUILD_CUDA_MODULE=ON，需确保基础镜像中已包含完整CUDA Toolkit
• 编译完成后可通过python3 -c "import open3d; print(open3d.__version__)"验证Python绑定是否安装成功

内容的提问来源于stack exchange，提问作者kronos