Polygon Threshold

Overview

The Polygon Threshold node filters a list of polygons by confidence score and size, passing through only those that meet the configured thresholds. It is commonly placed after an AI model node to remove low-confidence or undersized detections before they reach downstream logic.

Use Polygon Threshold when:

• You need to discard false-positive detections by setting a minimum confidence score.
• You want to ignore small, irrelevant objects by requiring a minimum polygon area.

💡 Tip

When both parameters are left at their defaults (0.0), the incoming polygon list passes through unchanged.

Inputs

Name	Type	Description
input	list[polygon]	Receives a list of polygons, typically produced by a model processor node.

Outputs

Name	Type	Description
output	list[polygon]	Returns only the polygons that satisfy both the probability threshold and minimum area requirements.

Parameters

Parameter	Default	Valid values	Unit	Tunable
Probability threshold	0.0	0.0 – 100.0	% (0–100)	✅
Minimum polygon area	0.0	0.0 – 100.0	% of image (0–100)	✅

Probability threshold

The minimum confidence score a polygon must have to pass through. The value is specified as a percentage — internally it is divided by 100 and compared against each polygon's probability. Polygons with a probability equal to or greater than the threshold are kept.

• Lower values keep more detections, including low-confidence ones. Useful during development or when you do not want to miss any objects, but may include more false positives.
• Higher values let only high-confidence detections pass through. Reduces false positives but may discard legitimate detections that fall below the threshold.

💡 Tip

This parameter complements the Score threshold set during object detection model training. The model's score threshold filters detections at inference time, while the Polygon Threshold node lets you fine-tune filtering afterwards without retraining.

Minimum polygon area

The minimum area a polygon must occupy, expressed as a percentage of the total image area. Polygons smaller than this threshold are discarded.

• Lower values allow smaller objects through. Set to 0.0 to disable area filtering entirely.
• Higher values let only large objects pass through. Useful for ignoring noise or small artifacts that are unlikely to be real detections.

Filtering logic

Each incoming polygon is evaluated against both thresholds in the following order:

1. Area check — If the polygon area is below Minimum polygon area, the polygon is discarded.
2. Probability check — If the polygon's confidence score is below Probability threshold, the polygon is discarded.
3. Polygons that pass both checks are included in the output.

INFO

Both checks must pass for a polygon to appear in the output. If a polygon fails either check, it is removed.

Example

Filtering low-confidence detections from an object detection model

A Faster R-CNN model detects objects with varying confidence. You want only detections the model is at least 50 % confident about, and you want to ignore any detection covering less than 1 % of the image.

Configuration:

• Probability threshold: 50.0
• Minimum polygon area: 1.0

Result: Only polygons with a confidence of 50 % or higher and an area of at least 1 % of the image are passed to the next node. All other detections are discarded.

Related links

Reference:

• Heatmap Threshold — similar thresholding node for heatmap data
• Basic information configuration — Score threshold — model-level score filtering configured during training