Real-Time Object Detection

A web-based real-time object detection application utilizing YOLO model for rock-paper-scissor detection. The trained weights are present as the game_weights.pt file.

A standalone version is a genreal purpose real-time detector written with TensorFlow.js and the Coco-SSD model. Detect objects through your webcam directly in the browser with no server required.

Demo

As for the rock-paper-scissors detector, install the requirements with:

pip install -r requirements.txt

Then run the live-time-detector.py normally:

python live-time-detector.py

[Make sure the env paths are set appropriately, which goes without saying.]

As for the standalone general purpose detector, just paste the open and run it on a modern browser as usual.

Supported Objects

The Coco-SSD model can detect 80 different object classes including:

• People and body parts
• Animals (cats, dogs, birds, etc.)
• Vehicles (cars, bikes, trucks, etc.)
• Household items (chairs, tables, TVs, etc.)
• Food items (pizza, banana, wine glass, etc.)
• And many more!

The other one is self-explanatory, I guess.

Privacy

• All processing happens locally in your browser
• No data is sent to external servers
• Camera feed is not recorded or stored

Technical Details

Rock-Paper-Scissors

• Model: YOLO11n (You Only Look Once v11 - nano variant)
• Framework: Ultralytics YOLO (PyTorch backend)
• Source: Custom-trained weights
• Captures frames from webcam using OpenCV

Standalone General Model

• Model: Coco-SSD (Common Objects in Context - Single Shot MultiBox Detector)
• Framework: TensorFlow.js
• Styling: Tailwind CSS
• Performance: Optimized for real-time detection with requestAnimationFrame

Dataset

Dataset: Rock-Paper-Scissors-SXSW
Publisher: Roboflow (via Universe)
URL: https://universe.roboflow.com/roboflow-58fyf/rock-paper-scissors-sxsw/dataset/14
Accessed: [10-08-2025]

Contributing

Feel free to fork this project and submit pull requests for improvements.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Troubleshooting

Camera not working?

• Ensure you've granted camera permissions
• Try using HTTPS (required for camera access on many browsers)
• Check if another application is using the camera

Poor detection performance?

• Ensure good lighting conditions
• Keep objects at a reasonable distance from the camera
• Try using a device with better processing power

Model loading slowly?

• It might take a while to load the model (usually <30 secs), please be patient.