Formating tensorflowjs object detection executeAsync output to prediction format

Question

I trained my own mobilenet ssd model, then I converted my model to tensorflow js but when the model does the prediction, the result is: enter image description here

T