deepstream-dli-course

Running the official NVIDIA Deep Learning Institute DeepStream course without any Jetson.

Setup

Download the official Docker image for DeepStream 7.0:

sudo docker pull nvcr.io/nvidia/deepstream:7.0-triton-multiarch

In case of running the container on WSL, follow these instructions. Afterwards, install specific X11 utils:

sudo apt install x11-xserver-utils

Run the container with ports and display overrides, GPU support and virtual mounts:

xhost +
sudo docker run -it --privileged --rm --name=docker --net=host --gpus all -e DISPLAY=$DISPLAY -e CUDA_CACHE_DISABLE=0 --device /dev/snd -v /tmp/.X11-unix/:/tmp/.X11-unix -v ~/:/opt/nvidia/deepstream/deepstream-7.0/my_home/ -p 8554:8554 nvcr.io/nvidia/deepstream:7.0-triton-multiarch

This will create in the working directory a link named my_home to the user home of the host, where the course files should be located and extracted.

Install the additional Python based applications:

./user_deepstream_python_apps_install.sh --version 1.1.11

Running Jupyter in Visual Studio Code

In Visual Studio Code install the Docker extension, then in the Docker sidebar right click on the executing container and select Attach Visual Studio Code. In the new window connected to the remote, install both the Python and Jupyter extensions as local extensions, which will manage the Jupyter kernel lifecycle.

Running Jupyter in web browser

Although Visual Studio Code is recommended for easier code edition during this course, Jupyter notebooks can be edited in a web browser.

Install the Jupyter runtime and webservice, then launch it:

pip install jupyterlab
pip install notebook
jupyter notebook --allow-root

To access the Jupyter web view without forwarding the Jupyter port and using a browser in host, install one in the container. For this, open a new terminal instance, get the container ID with docker ps, and attach it with docker attach <container-id>. Then run:

add-apt-repository -y ppa:savoury1/chromium
apt update
apt install -y chromium-browser
chromium-browser --no-sandbox

In the web browser, enter the URL indicated by the Jupyter server console output.

Video output

When running tests that display image, open VLC and in Media>Open Network Stream... introduce:

rtsp://localhost:8554/ds-test