Hand-pose-detection: 'tfjs-webgl' backend returns NaN values

hand-pose-detection: ‘tfjs-webgl’ backend returns NaN values.

I’m setting the hand-pose-detection model following the https://storage.googleapis.com/tfjs-models/demos/hand-pose-detection/index.html?model=mediapipe_hands example

In my code, setting the mediapipe runtime works fine, although setting the ‘tfjs-webgl’ runtime returns NaN values of x,y coordinates.

Detector initialization:

detectorConfig = {
        runtime: 'tfjs',
        modelType: 'full',
        maxHands: 1
      handDetector = await handPoseDetection.createDetector(model, detectorConfig)

Estimate hand data:

try {
      handData = await handDetector.estimateHands(sourceElement, 
        {flipHorizontal: isFlipHorizontal}         // flipHorizontal: false, if we want the result to flip horizontaly. Defaults is set to false.
    } catch (error) {
      handDetector = null

Logged detector output:

I would appreciate any suggestion on how to fix this issue. Maybe I’m missing some other detector configuration or setting?

I am having the same issue !
Using the following packages:

"@mediapipe/hands": "^0.4.1646424915",
    "@tensorflow-models/hand-pose-detection": "^2.0.0",
    "@tensorflow-models/handpose": "^0.0.7",
    "@tensorflow/tfjs-backend-webgl": "^4.2.0",
    "@tensorflow/tfjs-converter": "^4.2.0",
    "@tensorflow/tfjs-core": "^4.2.0",

the model returns NaN values !