Ubuntu 18.04上DeepStream安装
<p>DeepStream主页：https://developer.nvidia.com/deepstream-sdk</p><p>DeepStream Development Guide：https://docs.nvidia.com/metropolis/deepstream/4.0/dev-guide/</p>
<p>DeepStream SDK API：https://docs.nvidia.com/metropolis/deepstream/4.0/dev-guide/DeepStream_Development_Guide/baggage/index.html</p>
<p>环境搭建：</p>
<p>1）要求</p>
<div id="wwpID0E0ED0HA" class="List_Bullet" style="margin-left: 54pt; text-indent: -18pt"><span class="WebWorks_Number" style="width: 18pt">• Ubuntu 18.04</span></div>
<div id="wwpID0E0DD0HA" class="List_Bullet" style="margin-left: 54pt; text-indent: -18pt"><span class="WebWorks_Number" style="width: 18pt">• Gstreamer 1.14.1</span></div>
<div id="wwpID0E0CD0HA" class="List_Bullet" style="margin-left: 54pt; text-indent: -18pt"><span class="WebWorks_Number" style="width: 18pt">• NVIDIA driver 418+</span></div>
<div id="wwpID0E0BD0HA" class="List_Bullet" style="margin-left: 54pt; text-indent: -18pt"><span class="WebWorks_Number" style="width: 18pt">• CUDA 10.1</span></div>
<div id="wwpID0E0AD0HA" class="List_Bullet" style="margin-left: 54pt; text-indent: -18pt"><span class="WebWorks_Number" style="width: 18pt">• TensorRT 5.1.5</span></div>
<p>2）执行nvidia-smi查询Nvidia驱动版本</p>
<p>Driver Version: 418.87.00</p>
<p>执行cat /usr/local/cuda/version.txt查询CUDA版本</p>
<p>CUDA Version 10.1.243</p>
<p>执行cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2查询cudnn版本</p>
<p>7.6.3</p>
<p>执行cat /proc/version查询ubuntu系统版本</p>
<p>18.04.1</p>
<p>Gstreamer安装：</p>
<p><code>apt-get install 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</code></p>
<p>验证Gstreamer：dpkg -l | grep gstreamer</p>
<p><img src="https://img2018.cnblogs.com/blog/1617605/201909/1617605-20190916165346677-1024701938.png"></p>
<p>TensorRT安装：参考https://docs.nvidia.com/deeplearning/sdk/tensorrt-install-guide/index.html#installing-tar</p>
<p>下载：https://developer.nvidia.com/nvidia-tensorrt-5x-download，需要先登录NVIDIA账号</p>
<p>tar xzvf TensorRT-5.1.5.0.Ubuntu-18.04.2.x86_64-gnu.cuda-10.1.cudnn7.5.tar.gz</p>
<p>sudo vim ~/.bashrc,添加：</p>
<p><img src="https://img2018.cnblogs.com/blog/1617605/201909/1617605-20190916161343774-173683500.png"></p>
<p>source ~/.bashrc</p>
<p>cd TensorRT-5.1.5.0/python/</p>
<p>sudo -H pip3 install tensorrt-5.1.5.0-cp36-none-linux_x86_64.whl</p>
<p>（若没安装pip，先安装pip3：sudo apt install python3-pip）</p>
<p>cd ../uff (计划将TensorRT与TensorFlow一起使用时，安装uff才是必要的)</p>
<p>sudo -H pip3 install uff-0.6.3-py2.py3-none-any.whl</p>
<p>which convert-to-uff</p>
<p>若提示错误：</p>
<p><img src="https://img2018.cnblogs.com/blog/1617605/201909/1617605-20190929170512028-197934059.png"></p>
<p>先安装protobuf：</p>
<p>下载：https://github.com/protocolbuffers/protobuf/releases</p>
<div class="line number5 index4 alt2"><span style="color: rgba(0, 0, 255, 1)">cd protobuf-3.9.2/</span></div>
<div class="line number5 index4 alt2"><span style="color: rgba(0, 0, 255, 1)"><code class="bash plain">.</code><code class="bash plain">/autogen</code><code class="bash plain">.sh</code></span></div>
<div class="line number6 index5 alt1"><span style="color: rgba(0, 0, 255, 1)"><code class="bash plain"></code><code class="bash plain">./configure --prefix=/usr/local/protobuf</code></span></div>
<div class="line number7 index6 alt2"><span style="color: rgba(0, 0, 255, 1)"><code class="bash functions">sudo make -j12</code></span></div>
<div class="line number8 index7 alt1"><span style="color: rgba(0, 0, 255, 1)"><code class="bash functions"></code><code class="bash plain">sudo make install</code></span></div>
<div class="line number9 index8 alt2"><span style="color: rgba(0, 0, 255, 1)"><code class="bash functions"></code><code class="bash functions">sudo ldconfig</code></span></div>
<div class="line number9 index8 alt2"><span style="color: rgba(0, 0, 255, 1)">sudo vi ~/.bashrc</span></div>
<div class="line number9 index8 alt2"><span style="color: rgba(0, 0, 255, 1)">添加：PATH=/usr/local/protobuf/bin:$PATH</span></div>
<div class="line number9 index8 alt2"><span style="color: rgba(0, 0, 255, 1)">source ~/.bashrc</span></div>
<p>cd ../graphsurgeon</p>
<p>sudo -H pip3 install graphsurgeon-0.4.1-py2.py3-none-any.whl</p>
<p>验证TensorRT</p>
<p>cd TensorRT-5.1.5.0/samples/sampleMNIST/</p>
<p>make</p>
<p>cd ../../bin</p>
<p>./sample_mnist，如下所示，则安装成功</p>
<p><img src="https://img2018.cnblogs.com/blog/1617605/201909/1617605-20190916165052032-2093492519.png"></p>
<p>安装其他依赖：</p>
<p>sudo apt install \<br> libssl1.0.0 \<br> libgstrtspserver-1.0-0 \<br> libjansson4</p>
<p>3）安装librdkafka</p>
<p>git clone https://github.com/edenhill/librdkafka.git</p>
<p>cd librdkafka/<br>git reset --hard 7101c2310341ab3f4675fc565f64f0967e135a6a<br>./configure<br>make<br>sudo make install<br>sudo mkdir -p /opt/nvidia/deepstream/deepstream-4.0/lib<br>sudo cp /usr/local/lib/librdkafka* /opt/nvidia/deepstream/deepstream-4.0/lib<br>4）安装DeepStream SDK</p>
<p>下载：https://developer.nvidia.com/deepstream-sdk</p>
<p>tar -xpvf deepstream_sdk_v4.0_x86_64.tbz2</p>
<p>cd deepstream_sdk_v4.0_x86_64</p>
<p>sudo tar -xvpf binaries.tbz2 -C /</p>
<p>sudo ./install.sh</p><br><br>
来源：https://www.cnblogs.com/walker-lin/p/11528243.html
頁:
[1]