vision.model

Syntax

vision.model(model, labels?, roi?, threshold?, enabled?, backend?)

Parameters

model

Source model returned by models.set. Its model file and metadata configure the requested ONNX detector.

labels Optional

Optional class-label source: one label, an array of labels, or null when labels are not supplied.

roi Optional

Optional region of interest: a four-value [x, y, width, height] array. Pass null when the method accepts a full-frame scan.

threshold Optional

Detection or matching threshold. Higher values require a stronger match.

enabled Optional

Optional boolean that turns this configured tool on or off. Use true to process it and false to keep its configuration without running it.

backend Optional

Optional inference backend selection. Use the documented vision backend constant for the installed Atlas build.

Return value

Returns model.

Remarks

Configures an ONNX Runtime detector from a source model returned by models.set and returns a configured live model value. Only model is required. labels may be one string, an array, or null; roi defaults to null for the full frame, threshold to 0.5, enabled to true, and backend to vision.auto (vision.gpu and vision.directml both select DirectML). Read/set enabled, overlay, roi, threshold; read labels, results, count, target. Keep the separate source model for name/path reads.

Examples

model objectSource = models.set("models/object_detector.onnx");
model objectDetector;
const scanRadius = 0.18;
const scanRoi = [0.5 - scanRadius, 0.5 - scanRadius, scanRadius * 2, scanRadius * 2];

event on_start() {
    objectDetector = vision.model(objectSource, "object", scanRoi, 0.50, true, vision.auto);
}

event on_frame() {
    ui.circle(0.5, 0.5, scanRadius, rgb.cyan, 2, false);
    target detectedObject = objectDetector.target;
    if (detectedObject == null) {
        ui.status("No object detected");
        return;
    }
    ui.rect(detectedObject.box.x, detectedObject.box.y, detectedObject.box.width, detectedObject.box.height, rgb.lime, 3, false);
    ui.status(string.format("Object {}", detectedObject.confidence));
}

This complete, checked example also appears in Recipes.