This directory contains example Python scripts that demonstrate how to use PyNode workflows programmatically without the UI.

A basic example that captures frames from your webcam and displays them in an OpenCV window.

Features:

• Camera capture at 30 FPS
• Real-time display using OpenCV
• Configurable resolution (default: 640x480)

Usage:

python examples/camera_workflow.py

Controls:

• Press q in the display window to quit
• Press Ctrl+C in terminal to quit

An advanced example that captures camera frames, performs real-time YOLO object detection, and displays the annotated results.

Features:

• Camera capture at 30 FPS
• YOLOv8 object detection (uses yolov8n.pt - nano model)
• Real-time bounding boxes and labels
• FPS counter
• Detection count display
• Configurable confidence threshold

Usage:

python examples/camera_yolo_workflow.py

Requirements:

• OpenCV (opencv-python)
• Ultralytics YOLO (ultralytics)
• First run will download the YOLOv8n model (~6MB)

Performance Notes:

• Default configuration uses CPU
• For GPU acceleration, change 'device': 'cpu' to 'device': 'cuda' in the code
• YOLOv8n is the fastest model but less accurate
• For better accuracy, try yolov8s.pt, yolov8m.pt, or yolov8l.pt

Controls:

• Press q in the display window to quit
• Press Ctrl+C in terminal to quit

Both examples include a custom DisplayNode that demonstrates how to create your own nodes. You can extend this pattern to create any custom processing nodes:

class MyCustomNode(BaseNode):
    display_name = 'My Node'
    icon = '⚡'
    category = 'function'
    input_count = 1
    output_count = 1
    
    def on_input(self, msg: Dict[str, Any], input_index: int = 0):
        # Process incoming message
        payload = msg.get(MessageKeys.PAYLOAD)
        
        # Do your processing
        result = self.process(payload)
        
        # Send output
        output_msg = self.create_message(payload=result)
        self.send(output_msg)

All examples follow this pattern:

1. Import modules - Import the workflow engine and node types
2. Register node types - Register all available node types with the engine
3. Create nodes - Instantiate nodes with configuration
4. Connect nodes - Wire nodes together to create the workflow
5. Start workflow - Begin execution
6. Run loop - Keep the script running
7. Stop workflow - Clean shutdown

• The first YOLO inference may take a few seconds to load the model
• Adjust the camera FPS based on your processing pipeline speed
• Use JPEG encoding (encode_jpeg: True) for better performance
• Monitor the FPS counter to optimize your pipeline
• For production use, add error handling and logging

Camera not found:

• Check your camera_index (try 0, 1, 2)
• Make sure no other application is using the camera
• On Linux, check camera permissions

Low FPS:

• Reduce camera resolution
• Use a faster YOLO model (yolov8n.pt)
• Enable GPU if available (device: 'cuda')
• Reduce camera FPS to match processing speed

YOLO errors:

• Install ultralytics: pip install ultralytics
• Model will auto-download on first run
• Check internet connection for model download