Guide to Load and Run Peoplenet Deepstream 7.0 Model on a DGPU Device

Reference Docs:

1. https://docs.nvidia.com/metropolis/deepstream/6.0/dev-guide/text/DS_docker_containers.html
2. https://docs.nvidia.com/metropolis/deepstream/6.0/dev-guide/text/DS_ref_app_deepstream.html
3. https://docs.nvidia.com/metropolis/deepstream/6.0/dev-guide/text/DS_sample_custom_gstream.html
4. https://learn.nvidia.com/en-us/training/self-paced-courses

Have a quick reading through the Documentation.

1. Pull the docker container:

$ docker pull [nvcr.io/nvidia/deepstream:7.0-gc-triton-devel](http://nvcr.io/nvidia/deepstream:7.0-gc-triton-devel)

1. Connect the webcam and follow the below commands :)

#To run the docker as root:
docker run -it --rm --net=host --gpus all -e DISPLAY=$DISPLAY \
--device /dev/video0 --device /dev/snd \
-v /tmp/.X11-unix/:/tmp/.X11-unix \
-v /etc/nvidia:/etc/nvidia \
-v /opt/nvidia/deepstream/deepstream-7.0/samples/configs:/config \
-v /opt/nvidia/deepstream/deepstream-7.0/samples/models:/models \
-w /opt/nvidia/deepstream/deepstream-7.0/samples/configs/deepstream-app \
[nvcr.io/nvidia/deepstream:7.0-gc-triton-devel](http://nvcr.io/nvidia/deepstream:7.0-gc-triton-devel)

1. Open another separate terminal and type these commands,

#To give access to the camera:
$ xhost +
$ gst-launch-1.0 v4l2src device=/dev/video0 ! videoconvert ! autovideosink

1. Inside the docker terminal, type this command to edit the deepstream model config file (source1_usb_dec_infer_resnet_int8.txt):

$ nano source1_usb_dec_infer_resnet_int8.txt

1. Changes made to the config file:

[source0]
enable=1
#Type - 1=CameraV4L2 2=URI 3=MultiURI
type=1
uri=/dev/video0
camera-width=640
camera-height=480
camera-fps-n=30
camera-fps-d=1
camera-v4l2-dev-node=0

1. Command to run the deepstream model inside the docker:

$ deepstream-app -c source1_usb_dec_infer_resnet_int8.txt

There you go! Now you can see the output.

Made with 🫶🏻 by Basil

Check out my medium guide here: medium.com

You can go through my notion website: notion.com