Search code examples
iosswiftobject-detectioncoremlmlmodel

Dealing with MLMultiArray from MLModel

I'm using a mlmodel in my project to detect objects. The input for model is Image (Color 300 x 300) and outputs are MultiArray (Double) and MultiArray (Double 4 x 1917 x 1). I did VNCoreMLRequest and I got the output is exact as mentioned above. Here, I need to show the object name and confidence to the user from MLMultiArray output. Now, the question is how to decode it, how to get object name and confidence from the MLMultiArray. I'm researching from past 3 days and no use.

Solution

  • The number 1917 tells me you're most likely using MobileNet + SSD.

    I wrote a number of blog posts that explain how this sort of thing works. Here is a blog post that explains how to do the decoding for MobileNetV2+SSDLite, but the same logic works for other versions of MobileNet and SSD as well: https://machinethink.net/blog/mobilenet-ssdlite-coreml/