Bring real-time AI object detection to your network with YOLO-World-L, a dynamic custom frontend, and the powerful Luxonis OAK 4 (RVC4) camera. This project demonstrates how to use a live camera or video source, update classes dynamically, and control confidence thresholds—all through a browser.
Now with image upload and search, you can drag and drop any image into the UI to detect objects using custom or preloaded classes. Get instant bounding box annotations without writing a line of Python.
Inspired by modern open-vocabulary models and the Luxonis ecosystem, this is edge AI made flexible.
EXAMPLE
Here’s a short video showing the Dynamic YOLO-World system running live on the Luxonis OAK 4 (RVC4).
The video demonstrates:
Real-time object detection (YOLO-World-L)
Live confidence threshold updates
Dynamic class selection via the frontend
Annotated bounding boxes updating on the fly
Instant object search from uploaded image
SETUP
1. Clone the example
This uses Git to clone the necessary files from the Luxonis examples repo:
git clone https://github.com/luxonis/oak-examples.git
cd oak-examples
2. Start python virtual environment
cd custom-frontend/dynamic-yolo-world
python3 -m venv venv
source venv/bin/activate
3. Install Frontend dependencies
Move into the frontend folder and install the Node packages:
cd frontend/ && npm i
npm run build
cd ..
This builds the browser-based DepthAI Viewer with support for dynamic class and threshold input.
4. Connect to the OAK 4 Camera
Make sure your OAK 4 PoE device is powered and connected to the same network as your computer.
If you haven’t already, install oakctl (the Luxonis command-line tool).
Connect to your camera:
oakctl list
oakctl connect <DEVICE_IP>
Run the app on the device:
oakctl app run .
This deploys the pipeline and frontend to your OAK 4.
USAGE
Once the app is running, open your browser and go to:
https://<DEVICE_IP>:9000/
You’ll see the DepthAI Viewer, where you can:
View real-time annotated video from the OAK 4
Input or modify class names (like “chair”, “cup”, “laptop”)
Adjust the confidence threshold (e.g., from 0.1 to 1.0)
WRAP-UP
With just a few commands and the power of the Luxonis DepthAI stack, you’ve created a robust, responsive AI detection system that runs entirely on the edge.
Let your system detect only the things you care about, and let it evolve with your needs, live.
TROUBLESHOOTING
oakctl must be run in a system terminal (like Command Prompt, Terminal, or PowerShell), not inside Visual Studio Code’s terminal, as that may cause issues detecting network devices.
Tip: To easily get the full path to your project folder:
Right-click on the yolo-world-example project folder in Finder (macOS), File Explorer (Windows) or IDE of your choice.
Select “Copy as Path” (Windows) or “Get Info” → copy the path (macOS).
Paste it into your terminal with:
cd <paste-path-here>
- Then run:
oakctl list
oakctl connect <DEVICE_IP>
oakctl app run .
