Deep Stream Ai落地--初体验

Deep Stream

介绍

NVIDIA的DeepStream SDK提供了一整套数据流分析工具包，透过智能视频分析（IVA）和多传感器的数据处理来感知情景和意识。
DeepStream应用程序框架具有硬件加速构建块，可将深层神经网络和其他复杂处理任务带入流处理管道。开发者只需专注于构建核心深度学习网络和IP，而不是从头开始设计端到端解决方案。
更多详情了解，移步官网介绍，nvidia-deepSteam

解决问题

• 快速开发Ai技能
• 快速部署Ai服务
• 提供本地部署
• 提供边端设备部署
• 提供远端部署
• 高吞吐量

主要特点

• 具有统一规范的sdk
  • 基于多传感器，音频，视频，图像整套的流分析工具
• 具有基于graph composer拖拽式的低代码编程
• 支持云原声k8s编排
• 适用视觉Ai场景
• 高吞吐量

整体流分析过程

• 应用架构
  

• 流程开发
  

• 低代码构建
  

重点模块

• 加速解码器Gst-nvvideo4linux2

  • 流数据可以通过 RTSP 通过网络或来自本地文件系统或直接来自摄像机。使用 CPU 捕获流。一旦帧进入内存，它们就会被发送到使用 NVDEC 加速器进行解码

• 流缓冲区 Gst-nvstreammux

  • 缓冲区批量数据帧进行推理时可以更好的利用硬件资源

• 推理引擎 Gst-nvinfer

  • 本地端进行TensorRT的Inference，使用的方法是GST-nvinfer，使用TensorRT加速推理时，会做网络层之间的优化，且建立好一个可以被直接调用的推理引擎，该推理引擎可以被直接序列化，下次重新调用该引擎时，直接反序列化即可

• 格式转换Gst-nvvideoconvert

  • 批量转换，批量输出

• 可视化

  • gst-nvdsosd 可以帮助你根据实际场景绘制你感兴趣的部分

积木搭建

• 根据上面整体的结构以及重点模块，我们可以结合Deep Stream SDK 来构建自己业务的pipline

安装

安装必要的依赖

[~]# apt install \
libssl1.0.0 \
libgstreamer1.0-0 \
gstreamer1.0-tools \
gstreamer1.0-plugins-good \
gstreamer1.0-plugins-bad \
gstreamer1.0-plugins-ugly \
gstreamer1.0-libav \
libgstrtspserver-1.0-0 \
libjansson4 \
gcc \
make \
git \
python3

nvidia驱动安装

• 下载：https://www.nvidia.com/Download/driverResults.aspx/179599/en-us

cuda toolkit 安装

• 下载：https://developer.nvidia.com/cuda-11-4-1-download-archive

deep stream 安装

• 下载gpu版本(需要账号)：https://developer.nvidia.com/deepstream-getting-started

$ sudo tar -xvf deepstream_sdk_v6.0.0_x86_64.tbz2 -C /
$ cd /opt/nvidia/deepstream/deepstream-6.0/
$ sudo ./install.sh
$ sudo ldconfig

• ./samples目录是参考示例

docker运行实例

• 基于gpu

说明	拉取命令
基础 docker（仅包含运行时库和 GStreamer 插件。可用作为 DeepStream 应用程序构建自定义 docker 的基础）	docker pull nvcr.io/nvidia/deepstream:6.0-base
devel docker（包含整个 SDK 以及用于构建 DeepStream 应用程序和图形编辑器的开发环境	docker pull nvcr.io/nvidia/deepstream:6.0-devel
安装了 Triton 推理服务器和依赖项的 Triton 推理服务器 docker 以及用于构建 DeepStream 应用程序的开发环境	docker pull nvcr.io/nvidia/deepstream:6.0-triton
安装了 deepstream-test5-app 并删除了所有其他参考应用程序的 DeepStream IoT docker	docker pull nvcr.io/nvidia/deepstream:6.0-iot
DeepStream 示例 docker（包含运行时库、GStreamer 插件、参考应用程序和示例流、模型和配置)	docker pull nvcr.io/nvidia/deepstream:6.0-samples

• 以下为镜像构建的dockerfile参考样例，允许用户自定义镜像

# Set CUDA_VERSION, example: 11.4.1
ARG CUDA_VERSION
# Use CUDAGL base devel docker
FROM nvcr.io/nvidia/cudagl:${CUDA_VERSION}-devel-ubuntu18.04

# Set TENSORRT_VERSION, example: 8.0.1-1+cuda11.4
ARG TENSORRT_VERSION
# Set CUDNN_VERSION, example: 8.2.1.32-1+cuda11.4
ARG CUDNN_VERSION

# Install dependencies
RUN apt-get update && \
      DEBIAN_FRONTEND=noninteractive      apt-get install -y --no-install-recommends \
      linux-libc-dev \
      libglew2.0 libssl1.0.0 libjpeg8 libjson-glib-1.0-0 \
      gstreamer1.0-plugins-good gstreamer1.0-plugins-bad gstreamer1.0-plugins-ugly gstreamer1.0-tools gstreamer1.0-libav \
      gstreamer1.0-alsa \
      libcurl3 \
      libcurl3-gnutls \
      libuuid1 \
      libjansson4 \
      libjansson-dev \
      librabbitmq4 \
      libgles2-mesa \
      libgstrtspserver-1.0-0 \
      libv4l-dev \
      gdb bash-completion libboost-dev \
      uuid-dev libgstrtspserver-1.0-0 libgstrtspserver-1.0-0-dbg libgstrtspserver-1.0-dev \
      libgstreamer1.0-dev \
      libgstreamer-plugins-base1.0-dev \
      libglew-dev \
      libssl-dev \
      libopencv-dev \
      freeglut3-dev \
      libjpeg-dev \
      libcurl4-gnutls-dev \
      libjson-glib-dev \
      libboost-dev \
      librabbitmq-dev \
      libgles2-mesa-dev libgtk-3-dev libgdk3.0-cil-dev \
      pkg-config \
      libxau-dev \
      libxdmcp-dev \
      libxcb1-dev \
      libxext-dev \
      libx11-dev \
      git \
      rsyslog \
      vim  \
      gstreamer1.0-rtsp \
      libcudnn8=${CUDNN_VERSION} \
      libcudnn8-dev=${CUDNN_VERSION} \
      libnvinfer8=${TENSORRT_VERSION} \
      libnvinfer-dev=${TENSORRT_VERSION} \
      libnvparsers8=${TENSORRT_VERSION} \
      libnvparsers-dev=${TENSORRT_VERSION} \
      libnvonnxparsers8=${TENSORRT_VERSION} \
      libnvonnxparsers-dev=${TENSORRT_VERSION} \
      libnvinfer-plugin8=${TENSORRT_VERSION} \
      libnvinfer-plugin-dev=${TENSORRT_VERSION} \
      python-libnvinfer=${TENSORRT_VERSION} \
      python3-libnvinfer=${TENSORRT_VERSION} \
      python-libnvinfer-dev=${TENSORRT_VERSION} \
      python3-libnvinfer-dev=${TENSORRT_VERSION} && \
      rm -rf /var/lib/apt/lists/* && \
      apt autoremove


# Install DeepStreamSDK using debian package. DeepStream tar package can also be installed in a similar manner
ADD deepstream-6.0_6.0.0-1_amd64.deb /root

RUN apt-get update && \
      DEBIAN_FRONTEND=noninteractive  apt-get install -y --no-install-recommends \
      /root/deepstream-6.0_6.0.0-1_amd64.deb

WORKDIR /opt/nvidia/deepstream/deepstream

RUN ln -s /usr/lib/x86_64-linux-gnu/libnvcuvid.so.1 /usr/lib/x86_64-linux-gnu/libnvcuvid.so
RUN ln -s /usr/lib/x86_64-linux-gnu/libnvidia-encode.so.1 /usr/lib/x86_64-linux-gnu/libnvidia-encode.so

deepstream-python-app

• 使用镜像测试

$ docker pull nvcr.io/nvidia/deepstream:6.0-samples 

• 环境安装

apt-get install -y python-gi-dev
apt install -y python3-gst-1.0
apt install -y python3-pip
pip3 install pyds -i https://pypi.tuna.tsinghua.edu.cn/simple

• 拉取代码包

git clone https://github.com/NVIDIA-AI-IOT/deepstream_python_apps.git

• 示例代码

def main(args):
    # Check input arguments
    if len(args) != 2:
        sys.stderr.write("usage: %s \n" % args[0])
        sys.exit(1)

    # Standard GStreamer initialization
    GObject.threads_init()
    Gst.init(None)

    # Create gstreamer elements
    # Create Pipeline element that will form a connection of other elements
    print("Creating Pipeline \n ")
    pipeline = Gst.Pipeline()

    if not pipeline:
        sys.stderr.write(" Unable to create Pipeline \n")

    # Source element for reading from the file
    print("Creating Source \n ")
    source = Gst.ElementFactory.make("filesrc", "file-source")
    if not source:
        sys.stderr.write(" Unable to create Source \n")

    # Since the data format in the input file is elementary h264 stream,
    # we need a h264parser
    print("Creating H264Parser \n")
    h264parser = Gst.ElementFactory.make("h264parse", "h264-parser")
    if not h264parser:
        sys.stderr.write(" Unable to create h264 parser \n")

    # Use nvdec_h264 for hardware accelerated decode on GPU
    print("Creating Decoder \n")
    decoder = Gst.ElementFactory.make("nvv4l2decoder", "nvv4l2-decoder")
    if not decoder:
        sys.stderr.write(" Unable to create Nvv4l2 Decoder \n")

    # Create nvstreammux instance to form batches from one or more sources.
    streammux = Gst.ElementFactory.make("nvstreammux", "Stream-muxer")
    if not streammux:
        sys.stderr.write(" Unable to create NvStreamMux \n")

    # Use nvinfer to run inferencing on decoder's output,
    # behaviour of inferencing is set through config file
    pgie = Gst.ElementFactory.make("nvinfer", "primary-inference")
    if not pgie:
        sys.stderr.write(" Unable to create pgie \n")

    # Use convertor to convert from NV12 to RGBA as required by nvosd
    nvvidconv = Gst.ElementFactory.make("nvvideoconvert", "convertor")
    if not nvvidconv:
        sys.stderr.write(" Unable to create nvvidconv \n")

    # Create OSD to draw on the converted RGBA buffer
    nvosd = Gst.ElementFactory.make("nvdsosd", "onscreendisplay")

    if not nvosd:
        sys.stderr.write(" Unable to create nvosd \n")

    # Finally render the osd output
    if is_aarch64():
        transform = Gst.ElementFactory.make("nvegltransform", "nvegl-transform")

    print("Creating EGLSink \n")
    sink = Gst.ElementFactory.make("nveglglessink", "nvvideo-renderer")
    if not sink:
        sys.stderr.write(" Unable to create egl sink \n")

    print("Playing file %s " % args[1])
    source.set_property('location', args[1])
    streammux.set_property('width', 1920)
    streammux.set_property('height', 1080)
    streammux.set_property('batch-size', 1)
    streammux.set_property('batched-push-timeout', 4000000)
    pgie.set_property('config-file-path', "dstest1_pgie_config.txt")

    print("Adding elements to Pipeline \n")
    pipeline.add(source)
    pipeline.add(h264parser)
    pipeline.add(decoder)
    pipeline.add(streammux)
    pipeline.add(pgie)
    pipeline.add(nvvidconv)
    pipeline.add(nvosd)
    pipeline.add(sink)
    if is_aarch64():
        pipeline.add(transform)

    # we link the elements together
    # file-source -> h264-parser -> nvh264-decoder ->
    # nvinfer -> nvvidconv -> nvosd -> video-renderer
    print("Linking elements in the Pipeline \n")
    source.link(h264parser)
    h264parser.link(decoder)

    sinkpad = streammux.get_request_pad("sink_0")
    if not sinkpad:
        sys.stderr.write(" Unable to get the sink pad of streammux \n")
    srcpad = decoder.get_static_pad("src")
    if not srcpad:
        sys.stderr.write(" Unable to get source pad of decoder \n")
    srcpad.link(sinkpad)
    streammux.link(pgie)
    pgie.link(nvvidconv)
    nvvidconv.link(nvosd)
    if is_aarch64():
        nvosd.link(transform)
        transform.link(sink)
    else:
        nvosd.link(sink)

    # create an event loop and feed gstreamer bus mesages to it
    loop = GObject.MainLoop()
    bus = pipeline.get_bus()
    bus.add_signal_watch()
    bus.connect("message", bus_call, loop)

    # Lets add probe to get informed of the meta data generated, we add probe to
    # the sink pad of the osd element, since by that time, the buffer would have
    # had got all the metadata.
    osdsinkpad = nvosd.get_static_pad("sink")
    if not osdsinkpad:
        sys.stderr.write(" Unable to get sink pad of nvosd \n")

    osdsinkpad.add_probe(Gst.PadProbeType.BUFFER, osd_sink_pad_buffer_probe, 0)

    # start play back and listen to events
    print("Starting pipeline \n")
    pipeline.set_state(Gst.State.PLAYING)
    try:
        loop.run()
    except:
        pass
    # cleanup
    pipeline.set_state(Gst.State.NULL)


if __name__ == '__main__':
    sys.exit(main(sys.argv))

• 运行代码

$ cd /opt/nvidia/deepstream/deepstream-6.0/deepstream_python_apps/apps/deepstream-test1 
$ python3 deepstream_test_1.py /opt/nvidia/deepstream/deepstream-6.0/samples/streams/sample_720p.jpg
Creating Pipeline 
 
Creating Source 
 
Creating H264Parser 

Creating Decoder 

 Unable to create NvStreamMux 
 Unable to create pgie 
 Unable to create nvvidconv 
Creating EGLSink 

Playing file /opt/nvidia/deepstream/deepstream-6.0/samples/streams/sample_720p.jpg 
Traceback (most recent call last):
  File "deepstream_test_1.py", line 261, in <module>
    sys.exit(main(sys.argv))
  File "deepstream_test_1.py", line 194, in main
    streammux.set_property('width', 1920)
AttributeError: 'NoneType' object has no attribute 'set_property'

此bug还未解决

• 样例说明

名称	说明
deepstream-imagedata-multistream
deepstream-imagedata-multistream-redaction
deepstream-nvdsanalytics
deepstream-opticalflow
deepstream-rtsp-in-rtsp-out
deepstream-segmentation
deepstream-ssd-parser
deepstream-test1	如何将 DeepStream 元素用于单个 H.264 流的简单示例：filesrc → decode → nvstreammux → nvinfer (primary detection) → nvdsosd → renderer
deepstream-test1-rtsp-out
deepstream-test1-usbcam
deepstream-test2	如何将 DeepStream 元素用于单个 H.264 流的简单示例：filesrc → decode → nvstreammux → nvinfer（主检测器） → nvtracker → nvinfer（二级分类器） → nvdsosd → 渲染器
deepstream-test3
deepstream-test4
runtime_source_add_delete

Deep Stream Pipline 架构设计

Deep Stream 是一个基于GStreamer
，并由其插件来组建的流水线的过程

• Gst-nvstreammux:
  用于从多个输入源形成一批缓冲区
• Gst-nvdspreprocess:
  用于对预定义的 ROI 进行预处理以进行初级推理
• Gst-nvinfer:
  基于TensorRT的推理引擎
• Gst-nvtracker: 对象跟踪去重
• Gst-nvmultistreamtiler:
  用于形成 2D 帧数据
• Gst-nvdsosd:
  使用生成的元数据在合成帧上绘制阴影框、矩形和文本

有关graph Composer使用

安装中出现的问题可能在这里可以找到

借鉴思路

• Pipline流水式
• 组件式开发
• 拖拽式编程，块状可视化(流程图中块可修改代码)
• Pipline配置化
• 缓冲区设计
• …