Object Detector

Quick Start

To get started:

• Choose a trained model from the Model to use dropdown
• Set threshold values to control detection sensitivity
• Send image path via msg.payload.image_path
• Receive detected objects with bounding boxes in msg.payload

Configuration

Model to use (required)

Select a pre-trained model from the dropdown menu. Models must be trained beforehand using the object detection trainer block.

Confidence Threshold (required)

Minimum confidence score (0-1) for detected objects. Default: 0.5

Note: Lower values detect more objects with potentially lower confidence.

Bbox Filter Threshold (for v1 and Low Accuracy)

IOU threshold (0-1) for filtering overlapping bounding boxes. Default: 0.5

Note: Higher values allow more overlapping boxes, lower values filter out more overlaps.

Common Input Format (All Algorithms)

msg.payload.image_path (string)

Relative path of the image file on shared storage.

Example: "images/street.jpg" or "photos/product.png"

Supported formats: .png, .jpg, .jpeg (case insensitive)

Output by Model Type

msg.payload contains an output field with the detection results.

Low Infra - Good Accuracy - v1

msg.payload.output is an array of detections with the format: [[x1, y1, x2, y2, confidence, class_id], ...]

Example: {"output": [[100, 150, 300, 400, 0.89, 2], [350, 200, 500, 450, 0.76, 0]]}

Low Infra - Low Accuracy

msg.payload.output is an array of detections with the format: [[x1, y1, x2, y2, confidence, class_id], ...]

Example: {"output": [[100, 150, 300, 400, 0.89, 2], [350, 200, 500, 450, 0.76, 0]]}

Low Infra - Good Accuracy - v2

msg.payload.output is an array of detections with the format: [[x1, y1, x2, y2, class_id, confidence], ...]

Example: {"output": [[100, 150, 300, 400, 2, 0.89], [350, 200, 500, 450, 0, 0.76]]}

Note: Coordinates are [top-left-x, top-left-y, bottom-right-x, bottom-right-y]. class_id is a numeric label index.

Multiple threshold configuration

Some configurations expose multiple threshold fields for tuning recall vs precision.

Example

Input (msg.payload)

{
  "image_path": "images/street.jpg"
}

Output (msg.payload)

{
  "output": [
    [100, 150, 300, 400, 0.89, 2],
    [350, 200, 500, 450, 0.76, 0]
  ]
}

Errors

When the block fails, it raises an error. Use a Catch block in your flow to handle failures and inspect the error payload.

Common mistakes

• Missing or wrong image path: msg.payload.image_path is required and must point to a file on shared storage.
• Threshold out of range: Threshold values must be between 0 and 1.
• No objects detected: Lower the confidence threshold, or verify the model supports the objects present in the image.

Best Practices

• Use high-quality, well-lit images for optimal detection accuracy
• Start with confidence threshold 0.5 and adjust based on results
• Lower threshold (0.3-0.4) for detecting more objects, higher (0.6-0.8) for more reliable detections
• Use Bbox Filter Threshold around 0.5 to balance between duplicate detections and missing close objects
• Ensure training data includes diverse examples of objects in various conditions
• Regularly retrain models as object appearances or environments change
• Always validate detection results in production applications