Ultralytics YOLOv8 as Frigate Detector in Docker on Jetson Nano

Frigate NVR (Network Video Recorder) on Jetson Nano (JP4) - using YOLOv8 as a detector.

Frigate officially supports Deepstack as one of their Object detectors but Deepstack isn't actively maintained and here's yet another one, lightstack (only focusing on object detection) and trained on YOLOv8 nano model.

Setup Docker and Docker Compose

0.Super Duper update and upgrade

sudo apt-get update && sudo apt-get upgrade -y

1.Change docker's default runtime to nvidia

sudo vim /etc/docker/daemon.json

{
    ...
    "default-runtime": "nvidia" <<< Add this line
}

sudo systemctl restart docker

2.Add current user to docker group

sudo usermod -aG docker $USER && newgrp docker

3.Download docker-compose

curl -x "http://192.168.1.125:2080" -fsSL https://github.com/docker/compose/releases/download/v2.24.5/docker-compose-linux-aarch64 -o docker-compose

In Myanmar, IDK why Great Firewall is blocking githubusercontent.com, so I need to use a proxy in the curl command's -x option to bypass GFW.

4.Install docker-compose

DOCKER_CONFIG=${DOCKER_CONFIG:-$HOME/.docker}
mkdir -p $DOCKER_CONFIG/cli-plugins
cp docker-compose $DOCKER_CONFIG/cli-plugins/
chmod +x $DOCKER_CONFIG/cli-plugins/docker-compose

---

Frigate Docker Image for Jetson Nano Frigate REF

0.Pull the Mosquitto and Frigate images

docker pull eclipse-mosquitto:latest
docker pull ghcr.io/blakeblackshear/frigate:stable-tensorrt-jp4

Ultralytics YOLOv8 for Jetson Nano Ultralytics REF

0.First follow the best practices

sudo nvpmodel -m 0 # Running in MAX_5A
sudo jetson_clocks
sudo apt update && sudo pip3 install jetson-stats
sudo reboot
jtop

1.Export .pt to .engine using Ultralytics image

Export .pt to .engine first if you're using with Nvidia devices for better performance.

docker run --rm -it --runtime nvidia --network host --shm-size 1gb --gpus all -v $(pwd)/config/model_cache:/home/app/models ghcr.io/minlaxz/lightstack-api:yolov8-jp4 bash

yolo settings runs_dir="/home/app/models/runs" datasets_dir="/home/app/models/datasets" weights_dir="/home/app/models/runs/weights"

# yolo task=detect mode=export model=./yolov8-nano-best.pt format=engine data=license-plates/data.yaml int8=True imgsz=640 device=0

# Since `int8` isn't supported in Ultralytics TRT version for this Jetson Nano, data.yml is unnecessary here.
yolo task=detect mode=export model=./yolov8-nano-best.pt format=engine imgsz=640 device=0

2.Running lightstack-api

docker run --rm -it --runtime nvidia --network host --ipc host -v $(pwd)/config/model_cache:/home/app/models ghcr.io/minlaxz/lightstack-api:yolov8-jp4

---

Citation

@software{yolov8_ultralytics,
  author = {Glenn Jocher and Ayush Chaurasia and Jing Qiu},
  title = {Ultralytics YOLOv8},
  version = {8.0.0},
  year = {2023},
  url = {https://github.com/ultralytics/ultralytics},
  orcid = {0000-0001-5950-6979, 0000-0002-7603-6750, 0000-0003-3783-7069},
  license = {AGPL-3.0}
}