deeplearning4j/deeplearning4j · error · IllegalStateException

Invalid confidence threshold: must be in range [0,1]. Got:

Error message

Invalid confidence threshold: must be in range [0,1]. Got: 

What it means

YoloUtils.getPredictedObjects() validates the confidence threshold used to filter YOLO detections. Deeplearning4j throws this IllegalStateException because a threshold outside [0,1] would make the confidence filtering meaningless (confidences are probabilities in [0,1]). It fails fast instead of returning empty or bogus detections.

Source

Thrown at deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/objdetect/YoloUtils.java:185

     * (before getting to the Yolo2OutputLayer) we have 13x13 grid cells (each corresponding to 32 pixels in the input
     * image). Thus, a centerX of 5.5 would be xPixels=5.5x32 = 176 pixels from left. Widths and heights are similar:
     * in this example, a with of 13 would be the entire image (416 pixels), and a height of 6.5 would be 6.5/13 = 0.5
     * of the image (208 pixels).
     *
     * @param boundingBoxPriors as given to Yolo2OutputLayer
     * @param networkOutput 4d activations out of the network
     * @param confThreshold Detection threshold, in range 0.0 (least strict) to 1.0 (most strict). Objects are returned
     *                     where predicted confidence is >= confThreshold
     * @param nmsThreshold  passed to {@link #nms(List, double)} (0 == disabled) as the threshold for intersection over union (IOU)
     * @return List of detected objects
     */
    public static List<DetectedObject> getPredictedObjects(INDArray boundingBoxPriors, INDArray networkOutput, double confThreshold, double nmsThreshold){
        if(networkOutput.rank() != 4){
            throw new IllegalStateException("Invalid network output activations array: should be rank 4. Got array "
                    + "with shape " + Arrays.toString(networkOutput.shape()));
        }
        if(confThreshold < 0.0 || confThreshold > 1.0){
            throw new IllegalStateException("Invalid confidence threshold: must be in range [0,1]. Got: " + confThreshold);
        }

        //Activations format: [mb, 5b+c, h, w]
        long mb = networkOutput.size(0);
        long h = networkOutput.size(2);
        long w = networkOutput.size(3);
        long b = boundingBoxPriors.size(0);
        long c = (networkOutput.size(1)/b)-5;  //input.size(1) == b * (5 + C) -> C = (input.size(1)/b) - 5

        //Reshape from [minibatch, B*(5+C), H, W] to [minibatch, B, 5+C, H, W] to [minibatch, B, 5, H, W]
        INDArray output5 = networkOutput.dup('c').reshape(mb, b, 5+c, h, w);
        INDArray predictedConfidence = output5.get(all(), all(), point(4), all(), all());    //Shape: [mb, B, H, W]
        INDArray softmax = output5.get(all(), all(), interval(5, 5+c), all(), all());

        List<DetectedObject> out = new ArrayList<>();
        for( int i=0; i<mb; i++ ){
            for( int x=0; x<w; x++ ){
                for( int y=0; y<h; y++ ){

View on GitHub (pinned to 4c22ac5fe4)

Solutions

1. Pass the threshold as a fraction in [0,1], e.g. 0.5 for 50%
2. Divide a percentage-style config value by 100 before calling
3. Clamp the value with Math.min(1.0, Math.max(0.0, conf)) before the call

Example fix

// before
List<DetectedObject> objs = YoloUtils.getPredictedObjects(priors, output, 50, 0.45);
// after
List<DetectedObject> objs = YoloUtils.getPredictedObjects(priors, output, 0.5, 0.45);

if (confThreshold < 0.0 || confThreshold > 1.0) throw new IllegalArgumentException("confThreshold must be in [0,1], got " + confThreshold);

boolean validThreshold(double t) { return !Double.isNaN(t) && t >= 0.0 && t <= 1.0; }

try {
    objs = YoloUtils.getPredictedObjects(priors, output, conf, nms);
} catch (IllegalStateException e) {
    if (e.getMessage().contains("confidence threshold")) {
        objs = YoloUtils.getPredictedObjects(priors, output, Math.min(1.0, Math.max(0.0, conf)), nms);
    } else throw e;
}

When it happens

Trigger: Calling YoloUtils.getPredictedObjects(boundingBoxPriors, networkOutput, confThreshold, nmsThreshold) with a confThreshold < 0.0 or > 1.0, e.g. 50 or -0.1.

Common situations: Passing a threshold expressed as a percentage (0-100) instead of a fraction; loading a threshold from config in the wrong scale; sign errors when tuning thresholds.

Understand the failure class

Background: "value must be between 0 and 1" / "out of range" / "must not be negative" errors: fixing range-validation failures across open-source libraries — this error's family across 42 libraries.

Related errors

• Classifier evaluation using ${classifierEval.getSimpleName()
• Invalid input: (x1,y1), top left position must have values l
• Cannot construct RecordReaderMultiDataSetIterator with no re
• Ratio value should be in range of 0.0 > X < 1.0
• totalExamples number should be positive value

AI-assisted analysis of deeplearning4j/deeplearning4j@4c22ac5fe4 (2026-09-07). Data as JSON: /api/errors/12fe20a0bebab018. Report an issue: GitHub.