Setup

This page details setup steps needed to start using JetNet

Docker Setup

JetNet comes with pre-built docker containers for some system configurations. If you have the disk space and there is an available container, this is a fast and easy option for getting started.

To use the container, first clone the github repo

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

Next, launch the docker container from inside the cloned directory

Jetson (JetPack 5.0.2)Jetson (JetPack 5.0.1)Desktop (NV driver 465.19.01+)

docker run \
    --network host \
    --gpus all \
    -it \
    --rm \
    --name=jetnet \
    -v $(pwd):/jetnet \
    --device /dev/video0 \
    -v /tmp/.X11-unix:/tmp/.X11-unix \
    -e DISPLAY=$DISPLAY \
    jaybdub/jetnet:l4t-35.1.0 \
    /bin/bash -c "cd /jetnet && python3 setup.py develop && /bin/bash"

docker run \
    --network host \
    --gpus all \
    -it \
    --rm \
    --name=jetnet \
    -v $(pwd):/jetnet \
    --device /dev/video0 \
    -v /tmp/.X11-unix:/tmp/.X11-unix \
    -e DISPLAY=$DISPLAY \
    jaybdub/jetnet:l4t-34.1.1 \
    /bin/bash -c "cd /jetnet && python3 setup.py develop && /bin/bash"

docker run \
    --network host \
    --gpus all \
    -it \
    --rm \
    --name=jetnet \
    -v $(pwd):/jetnet \
    --device /dev/video0 \
    -v /tmp/.X11-unix:/tmp/.X11-unix \
    -e DISPLAY=$DISPLAY \
    jaybdub/jetnet:x86-21.05 \
    /bin/bash -c "cd /jetnet && python3 setup.py develop && /bin/bash"

This will mount the current directory (which should be the jetnet project root) at /jetnet inside the container. Most data downloaded when using JetNet is stored in the data folder. So assuming you use JetNet command line tools from /jetnet inside the container, the data will persist upon container restart.

Note, this command assumes you have a USB camera at /dev/video0. Please adjust the command accordingly.

Building docker containers

You may want to build the containers yourself, if you have additional dependencies, or need to use a different base container. Below are the commands we use to build the pre-made containers. Check the GitHub repo docker files for more details.

docker build -t jaybdub/jetnet:l4t-34.1.1 -f $(pwd)/docker/l4t-34.1.1/Dockerfile $(pwd)/docker/l4t-34.1.1

Manual Setup

If there is not a container available for your platform, or you don't have the storage space, you can set up your system natively.

• Install TensorRT, PyTorch, OpenCV and Torchvision (please refer to external instructions)

• Install miscellanerous dependencies

  pip3 install pydantic progressbar python3-socketio uvicorn starlette
  

• Install torch2trt

  pip3 install git+https://github.com/NVIDIA-AI-IOT/torch2trt.git@master
  

• Install YOLOX (required for jetnet.yolox)

  git clone https://github.com/Megvii-BaseDetection/YOLOX
  cd YOLOX
  python3 setup.py install
  cd ..
  

• Install EasyOCR (required for jetnet.easyocr)

  pip3 install git+https://github.com/JaidedAI/EasyOCR.git@v1.5.0
  

• Install TRTPose (required for jetnet.trt_pose)

  pip3 install git+https://github.com/NVIDIA-AI-IOT/trt_pose.git
  

• Install JetNet

  git clone https://github.com/NVIDIA-AI-IOT/jetnet
  cd jetnet
  python3 setup.py develop
  

Currently we exclude jetnet.mmocr from manual setup. For now, please reference the dockerfile in the GitHub repo if you wish to use these models.