As it says on the tin.

Make sure you have docker and nvidia-docker installed and attach a V4L2 compatible camera and note its device id. Start a docker container using

$ docker run --mount source=isaac-sdk-build-cache,target=/root -v <path to project directory>:/workspace -w /workspace --gpus=all --device <path to camera eg: /dev/video2> --net=host -it firekind/isaac:2020.2-deepstream-5.0.1-devel /bin/bash

then, download the models.

/workspace$ ./download-models.sh

This will download the models into the models/yolo directory. Then, compile the custom deepstream yolo plugin.

$ cd lib
$ export CUDA_VER=10.2
$ make

Edit the device_id under the config section of app/graphs/detector.app.json file to the device id of your setup. Then, run:

$ bazel run //apps:detector

and open localhost:3000 on the browser to see the results.

Note: By default, the graph uses yoloV3 tiny. To use yoloV3, edit the config-file-path property of the nvinfer element in app/graphs/detector.app.json to app/configs/yolov3-config.txt